Journal of Computing & Biomedical Informatics

An Ensemble Approach for Firewall Log Classification using Stacked Machine Learning Models

2025-03-08T12:02:20+00:00

Firewall logs are still challenging to evaluate, while being important data sources. Machine Learning has become a popular technology for creating strong security measures because of their ability to react quickly to complicated attacks. Firewall logs generate high-volume, complex, and often imbalanced data, where malicious activities are rare compared to normal traffic. The challenge is further compounded by the dynamic nature of cyber threats and the presence of noise or redundant information in the logs. In this research, a stacking classifier called Decision Tree Classifier + Bagging Classifier (DB) for Firewall logs is proposed using the ensemble machine learning models. A comparison is performed to evaluate the classifier's overall performance based on F1-score, accuracy, precision, and recall. A firewall that was set up with Snort and TWIDS had its logs taken. The 65532 occurrences of the receiving log record include a total of 12 attributes. Creating multi-class machine learning models that can analyze the firewall logs dataset and classify the necessary actions in response to learned classes as "Reset-both," "Allow," "Deny," or "Drop". For assessment, a variety of machine learning methods have been used, such as Random Forest, K-Nearest Neighbor, Logistic Regression, and AdaBoost Classifier. The experiment's 99.89% accuracy rate for the proposed model using stacking classifier DB is an interesting interpretation of the findings. However, the high accuracy rates produced as compared to other algorithms show that the recommended points were crucial in increasing the firewall classification rate.

Quantum-Resistant Merkle Trees Enhancing Data Integrity with Post-Quantum Cryptography and Zero-Knowledge Proof

2025-04-27T16:49:42+00:00

Exponential advancements in quantum computing threaten existing cryptographic structures, including Merkle Trees, due to their dependence on classical hash functions and public-key encryption schemes. The paper presents QRMT as a new cryptographic structure that implements zk-STARKs along with lattice-based cryptography and hash function randomization to achieve improved security and better performance. Benchmarks demonstrate that QRMT reduces proof generation time by 28–32% compared to classical Merkle Trees under Grover’s algorithm attacks, while maintaining logarithmic-scale verification efficiency. The QRMT utilizes a hash selection strategy that consists of SHAKE-256 Blake3 and Poseidon hash functions, which protect against Grover’s algorithm attacks. The metadata encryption measures security through Kyber1024, which uses lattice-based public-key encryption to replace RSA and prevent attacks using Shor’s algorithm. Kyber1024 generates keys in ~0.005 ms, which is 75 ms faster than RSA-4096’s. The zk-STARK-verified process allows for trustless and extensive proof verification while protecting confidential information. Our proof-of-concept instance maintains efficient behavior because proof creation and verification times grow at less than a logarithmic rate while the data collection expands. This framework creates quantum resistance for blockchain security, which enables distributed secure systems and establishes new cryptographic technology options.

Software Development Empowered and Secured by Integrating A DevSecOps Design

2025-03-11T15:32:17+00:00

This made the development of software grow fast, injecting speed and agility in the processes of delivery of software, but integrating security into these high-speed environments has remained a challenge. The solution to this problem comes through the adoption of a methodology known as DevSecOps, encompassing security at each step in the lifecycle of software development. It explored the adoption and value of DevSecOps, concentrating more on automation, vulnerability detection, and continuous security testing. It outlines a comprehensive review of available literature on the topic, with a special focus on the leading tools in this list, namely Static Application Security Testing (SAST), Software Composition Analysis (SCA), and Dynamic Application Security Testing (DAST). The paper will go on to discuss real examples of DevSecOps implementation and follow that up with a discussion of emerging trends, such as machine learning, cloud-native security, and zero-trust models. The study depicts the fact that, though DevSecOps has not matured as a concept yet, its adoption is at a very critical phase in building secure, efficient, and resilient software systems.

Ethical Considerations in Utilizing Machine Learning for Depression and Anxiety Detection in College Students

2025-03-18T18:41:30+00:00

Depression and Anxiety are two of the most common mental disorders that are happening nowadays worldwide. This systematic review paper investigates the application of neural network-based machine learning techniques in assessing, identifying, and diagnosing depression and anxiety, which are two worldwide mental health problems. These approaches can range from many different types of neural network architectures like classical supervised learning methods to unsupervised approaches with recent deep models; if a systematic literature review of studies conducted in the last five years is carried out using databases like Science Direct and PubMed, these approaches can show promising results in tasks like clinical data analysis, biomarker identification, and personalized treatment plan creation. This makes it possible to explain the 91.08% accuracy ratio that he reported in his study, which gave rise to a comprehensive confusion matrix and analysis of the classification report for the employed neural network model (Level 6). There are still difficulties in correctly identifying cases of depression (class 1) despite relatively high recall and precision in non-depressive cases (class 0). This raises possible areas for improvement, particularly with the current class imbalances. This review paper can serve as a valuable source of information for all interested in neural network-based machine learning and how it may be used to address depression and anxiety, thereby providing insight into the future that will likely change mental treatment models.

Real-Time Text Document Classification Using Fully Connected Neural Network

2025-04-30T20:52:31+00:00

Automated classification of text documents stands crucial in modern times because of the rising digital information volumes. The substantial amount of textual data across different industries gets better handled through automated document classification, which helps both retrieval and analysis, and organization of massive data collections. The system enables fast classification of documents, which leads to better decisions which resulting in improved productivity and simplified organizational processes. The proposed system implements a complete automated text document classification through deep learning (DL) methodologies. The data gets saturated by first removing special characters, together with common non-alphanumeric characters. Our proposed Fully Connected Neural Network (FCNN) receives the pre-processed database that has undergone tokenization. The proposed methodology achieved maximum accuracy at 99%. The method demonstrates strong reliability in processing real-time text data classification operations.

Unveiling 6G Networks: Innovations, Challenges, and Future Research

2025-03-04T03:30:52+00:00

With the advent of fifth-generation (5G) wireless communication technology, several intelligent applications are being integrated into various domains. However, 5G specifications fall short of meeting the demands of emerging technologies such as connected autonomous vehicles, artificial intelligence (AI) / cloud integration, Smart Grid 2.0, collaborative robots, Industry 5.0, digital twins, extended reality, and hyper-intelligent healthcare. These cutting-edge applications require enhanced technical capabilities, including higher data rates, greater network capacity, ultra-low latency, improved reliability, efficient resource allocation, expanded bandwidth, and optimal energy efficiency per bit. Since existing 5G technology does not fully address these evolving requirements, research and development efforts must pivot toward sixth-generation (6G) wireless technologies to bridge the existing gap. This study comprehensively unveils 6G innovations, exploring its key technological advancements, including ultra-low latency, enhanced data rates, and improved energy efficiency. Additionally, it identifies critical challenges such as security attacks, data privacy risks. The study also highlights potential research directions to address these challenges and ensure the successful deployment of 6G.

MalwareVison: A Deep Learning-Driven Approach For Malware Classification

2025-05-03T08:14:07+00:00

The fast propagation of malware across the internet requires the development of advanced classification and detection techniques. Traditional signature-based detection malware methods often fail to identify new and obfuscated variants which demand advanced machine learning-based solutions. We propose MalwareVision, a framework based on deep learning for the classification of malware samples. The model was trained on the Malimg dataset comprising images of 9,339 malware images across 25 families and evaluated based on accuracy, precision, recall, and F1-score metrics. We observe that the model achieved an impressive accuracy of 95.09% in both the training and testing datasets and that the precision and recall values remained high for most malware families. The results highlight the effectiveness of deep learning-based Convolutional Neural Network (CNN) for malware classification. The proposed MalwareVision framework offers a scalable, automated solution for malware classification, contributing to the advancement of AI-driven cybersecurity defenses.

Multispectral Approach for Wheat Yield Estimation Using Deep Learning

2024-12-30T13:11:28+00:00

Automation is becoming increasingly vital across various professions and domains, including agricultural practices. Remote-sensing-based wheat yield estimation has emerged as a superior alternative to traditional yield prediction methods. Historically, wheat yield measurement involved labor-intensive and time-consuming destructive sampling techniques. However, accurate and timely yield forecasts are pivotal for decision-making processes such as crop harvesting plans, milling, marketing, and forward selling strategies, thereby enhancing the efficiency and profitability of the global wheat sector. Presently, producers or productivity officers, often funded by mills, rely on destructive or visual sampling techniques to assess wheat production during the growing season. There's a growing demand for swift and efficient problem-solving methods. Consequently, the adoption of machinery for wheat cultivation has surged, aiming to lower production costs, reduce labor demands on farmers, and enhance harvest efficiency. Although not extensively compared, existing techniques for estimating agricultural output typically employ regression models relying on specific forecasting factors. This study aims to illustrate and compare the effectiveness of utilizing satellite earth observation data for monitoring agriculture, particularly in wheat production. Multiple regression models are compared, utilizing various predictor variables. The study incorporates wheat yield estimation techniques, such as regression models, time series analysis of vegetation indices, remote sensing, phenology measurements, and normalized difference vegetation index (NDVI). Artificial intelligence algorithms, including Random Forest and ordinary least squares, are employed to develop a suggested approach that accurately correlates ground-measured data. This research introduces a novel wheat yield estimation technique, which significantly improves forecasting accuracy and holds promise for enhancing decision-making processes in wheat farming practices.

Blockchain in the Digital Age: Challenges, Opportunities, and Future Trends

2025-02-07T09:57:52+00:00

Blockchain technology offers a massive network with built-in security features that encompass cryptography, decentralization, and consensus, which foster trust in transactions. IoT further looks as an emerging in finance and security that are the application of blockchain. The basic need for every blockchain consumer is the first to prioritize data confidentiality, integrity, and availability. In the era of 2025, trust is a necessary part of security for third parties, which handle the privileges of private and public. The mentioned advantages and disadvantages motivated us to provide an advancement and comprehensive study regarding the applicability of blockchain technology. This paper focuses on blockchain security issues for blockchain and sorts out the security risks in six layers of blockchain technology by comparing and analyzing existing security measures. The text also investigates and describes various security threats and obstacles associated with implementing blockchain technology, fostering theoretical inquiry and the creation of strong security protocols in current and forthcoming distributed work settings.

Integrated Power Combining and Amplification for Compact S-Band Radar Transmitters

2024-12-19T10:17:05+00:00

This paper deals with one of the major problems in high power S band radar transmitters by discussing the design, simulation, fabrication, and validation of a compact, high performance Wilkinson type power divider with amplifier impedance matching network. Till now, traditional Wilkinson Power Dividers (WPDs) tend to focus on isolation and return loss but are poor in amplitude balance throughout the wide frequency ranges especially in high power and space limited applications. The architecture proposed is applicable for a 1 kW S-band radar transmitter and focuses on minimal amplitude imbalance (≤0.75 dB), compact, return loss better than −12 dB. When simulations were conducted using Advanced Design System (ADS) software, performance was shown to be strong with insertion loss between −12.18 dB and −12.50 dB, return loss more than −18.65 dB, and isolation more than −25 dB in the bandwidth, 2.7–3.0 GHz. The design was built using a Rogers RT/Duroid 5870 substrate, the design utilized a 75W PH2731-75L RF transistor with optimized impedance matching to have a high power dissipation. Simulations came relatively close to measurement results, which confirmed the effectiveness of the design. This work helps to build compact and efficient RF front end modules for radar that can be upgraded from the insertion loss and future applications wise.

AI-Enhanced Bioactive 3D-Printed Scaffolds for Tissue Regeneration: Innovations in Healing and Functional Additives

2025-02-14T10:28:09+00:00

Bioactive 3D-printed scaffolds have revolutionized tissue engineering and regenerative medicine by enabling precise fabrication of biomimetic structures that promote cell adhesion, proliferation, and differentiation. However, significant challenges remain, particularly in optimizing scaffold composition, bioactive additive integration, and long-term stability for clinical applications. This review provides a systematic analysis of recent advancements in AI-driven bioactive scaffolds and their role in personalized regenerative medicine. A comparative evaluation of major 3D printing techniques—Fused Deposition Modeling (FDM), Stereolithography (SLA), Selective Laser Sintering (SLS), and Direct Metal Laser Sintering (DMLS)—is presented, focusing on resolution, material compatibility, and bioactive additive incorporation. Additionally, we analyze key bioactive agents (growth factors, nanoparticles, peptides, and natural polymers) and their effects on biocompatibility, mechanical strength, and therapeutic efficacy. AI-powered optimization techniques, including machine learning-based scaffold design, computational modeling, and predictive analytics, are emerging as transformative solutions for improving scaffold architecture, drug delivery systems, and patient-specific applications. Despite significant progress, major challenges persist, including standardization in scaffold fabrication, long-term in vivo validation, and regulatory approval hurdles. Addressing these scientific and regulatory challenges is essential for the successful clinical translation of bioactive scaffolds. This review highlights the need for interdisciplinary collaboration to advance AI-assisted scaffold engineering and establish personalized treatment strategies for next-generation regenerative medicine.

Topic Modeling Empowered by a Deep Learning Framework Integrating BERTopic, XLM-R, and GPT

2025-04-29T18:53:37+00:00

Topic modeling facilitates the identification of hidden themes and patterns in large text collections. It enables a thorough investigation of the messages contained in texts. Topic modeling is a popular research subject, with several translations already being investigated, including English and Arabic. However, there is a need for more research into low-resource languages, including Urdu. In this study, we propose using the BERTopic, XLM-R, and GPT frameworks on Urdu text. The proposed approach, which includes fine-tuned BERT, XLM-R, and GPT models, aims to capture the contextual nuances and grammatical intricacies of Urdu text. In this investigation, we used existing Urdu textual data. We evaluated the performance of our proposed approaches to existing techniques such as LDA and NMF utilizing coherence and diversity measures. The results show that our proposed strategy outperforms existing methods, with an average coherence improvement of 0.05 and a diversity score of 0.87. These findings demonstrate the efficacy of the proposed approach in extracting significant topics from Urdu texts, hence assisting scholarly endeavors in comparative studies of Urdu translations. Integrating real-time Urdu topic modeling into social media and news monitoring systems can help in trend analysis, misinformation detection, and sentiment-aware content moderation. Another practical application is the incorporation of topic modeling in Urdu search engines and recommendation systems, improving information retrieval for Urdu-speaking users.

Personality Prediction of the Users Based on Tweets through Machine Learning Techniques

2025-01-03T07:48:00+00:00

With the advancement of interpersonal organizations, the massive rich data from Social platforms such as Facebook, YouTube, Instagram, and Twitter supply determining information about Social communications and human manners. An assortment of approaches has been created to characterize clients' characters dependent on their social exercises and language use propensities. Specific methodologies vary to various AI calculations, information sources, and capabilities. The Informal community application records the enormous measure of users' conduct communicated in different exercises like preferences, notices, posts, remarks, photographs, labels, historical textual features, tweets, user profiles, and offers. Different analysts use numerous old Machine Learning calculations in establishing their models. This examination attempts to execute a few extreme learning designs to see the correlation by exhaustive investigation strategy through the exact results. We inspect the presence of plans of interpersonal organizations and semantic attributes comparative with character connections utilizing the myPersonality project dataset. The investigation also looks at two AI models and plays out the link between every one of the feature sets and personality traits. The outcomes for the forecast exactness show that regardless of whether tried under a similar dataset, the character expectation framework based on the Logistic Regression classifier outflanks the standard for all the included sets, prediction accuracy of 98.9%. The best prediction accuracy of 99.8% is gained by using the Random Forest classifier.

A Comprehensive Review on the Role of AI in Phishing Detection Mechanisms

2025-02-13T12:08:26+00:00

The integration of Artificial Intelligence (AI) in various domains, such as healthcare, education, business, and generative applications, is thoroughly reviewed in this paper emphasizing the transformative and detective power of AI. The operational efficiency of the applications has been greatly increased by AI-driven innovations like automation, personalized healthcare solutions, and improved decision-making optimization. In addition to providing a critical analysis of AI's advantages and disadvantages, the review synthesizes findings from several studies and addresses ethical concerns, data privacy issues, bias, and security risks related with generative AI technologies. The implications of AI-driven disinformation and the changing regulatory environment are two promising new insights this paper offers. Even with significant advancements, there are still issues that need to be addressed, especially with regard to managing the social effects of AI, creating uniform ethical frameworks, and guaranteeing data quality. This study adds to the body of literature by highlighting the necessity of strong mitigation techniques, interdisciplinary cooperation, and ethical AI governance to promote responsible AI deployment. Recommendations for future work, including the need to develop robust strategies of alleviation, ethical instructions and interdisciplinary cooperation to ensure that AI technology is successfully implemented.

Evaluation of Different Irrigation Settings to Maximize Water Productivity

2024-12-30T13:26:40+00:00

This study evaluates the impact of different irrigation systems on maize crop performance, focusing on plant height, stem diameter, grain yield, and water productivity. The results demonstrate that drip irrigation is the most effective system, producing the tallest plants (254.6 cm), the largest stem diameter (17.9 mm), and the highest grain yield (6,728.3 kg/ha). It also achieved the greatest water use efficiency at 15.62 kg/m³, with minimal water wastage. The sprinkler irrigation system also performed well but with slightly lower metrics compared to drip irrigation. Conversely, furrow and flood irrigation systems resulted in shorter plants, smaller stem diameters, and lower yields, with significantly lower water productivity due to higher water usage and wastage. The study underscores the advantages of drip irrigation in enhancing maize growth and yield while conserving water, highlighting its potential as a sustainable agricultural practice.

Crime Forecasting Using Data Analytics in Pakistan

2025-03-24T08:25:49+00:00

According to the 2017 Census, Lahore has a population of 11,126,285, with a police-to-citizen ratio of 1:413—significantly higher than that of many other major metropolitan cities around the world. Given these limited resources, there is an urgent need to develop effective strategies to reduce and control crime in a city like Lahore. One viable solution lies in the ability to predict crimes using historical data. Machine learning and data mining techniques can play a crucial role in forecasting criminal activity and identifying common crime patterns. In this study, we applied regression analysis to a real 15-Lahore crime dataset provided by the Punjab Safe City Authority (PSCA) to predict various crime attributes, including location, time, and type. Additionally, association data mining and clustering techniques used to discover frequent patterns and categorize crimes into distinct clusters. The primary goal of this research is to enable the efficient use of existing resources by employing predictive analytics to support proactive crime prevention.

A Framework for Sarcasm Detection Incorporating Roman Sindhi and Roman Urdu Scripts in Multilingual Dataset Analysis

2025-04-18T19:15:02+00:00

Sarcasm detection is imperative for successful real-time sentiment analysis in the pervasive social web. Detection of sarcastic tones expressed through text that impart bitter, satirical, or mockery expressions, remarks, or derision in Natural Language Processing (NLP) is problematic to handle for humans; making it automated is even more arduous. This work aims to propose a sarcasm detection framework tflexihat optimizes a sentiment analysis system by correctly detecting sarcastic text messages for resource-poor languages in multilingual datasets. The techniques developed to date are inadequate and require precise training data. Therefore, we propose neural networks and deep learning-based models that focus on contextual information utilizing different word embedding techniques, and we further propose a framework for multilingual sarcasm detection resources for low-resource languages such as Roman Sindhi and Roman Urdu. With this sarcasm-aware framework, individuals with limited English proficiency will be better equipped to engage on social media using sarcastic tones, emojis, and creative linguistic variations in a multilingual textual data analysis.

Decision Making Framework for Phishing Incident Response Using Intuitionistic Fuzzy Trapezoidal Preference Relations

2025-03-10T08:11:05+00:00

Decision-making often involves uncertainty, requiring precise methodologies to ensure accuracy and reliability. This paper presents an intuitionistic fuzzy trapezoidal preference relation (IFTrPR)-based decision-making framework that integrates multiplicative consistency for improved priority weight assessment. The proposed approach determines intuitionistic fuzzy trapezoidal priority weight vectors and ranks alternatives using the technique for order preference by similarity to the ideal solution (TOPSIS). To enhance the consistency of priority weights, a Linear Decision Model (LDM) is employed, effectively capturing decision-makers’ perceptions. Additionally, Model 1 is introduced to compute priority weights based on intuitionistic fuzzy trapezoidal numbers (IFTNs) across various alternatives. The integration of fuzzy logic and optimization techniques strengthens the framework’s ability to handle complex decision-making problems. A comparative analysis with hierarchical fuzzy systems (HFS) demonstrates that the proposed method enhances accuracy and reliability in priority weight assessment. Furthermore, the study provides a systematic approach to handling linguistic variables in decision-making, particularly in the representation of membership (MS) and non-membership.

Sentiment Analysis of Urdu Text using Hybrid Deep Learning Techniques

2025-05-04T05:46:35+00:00

Sentiment Analysis (SA) has become the rising topic of research in data mining. There is a massive increase in the use of the Internet and business e-commerce applications. With the excess amount of text content on the internet, Sentiment Analysis has begun to attract the attention of more people. Sentiment analysis of single sentence referred as short text, is a very tricky task because single sentence does not have much contextual information. Furthermore, not much work has been done in field of sentiment analysis of Urdu text because a very few amounts of data and recourses of Urdu text are available publicly. The purpose of this research is to contribute in the field of sentiment analysis of Urdu text. Mostly in Natural Language Processing (NLP), tasks words are the main focus. This research proposed a deep learning techniques combination of Convolutional Neural Network (CNN) with Long Short-Term Memory (LSTM) for the Sentiment Analysis of Urdu Text. In this technique, word embedding is performed using fastext API which poses a high dimensionality (300 dimensions). Long Short-Term Memory helps capturing the long-term dependencies and minimizes the loss of local information. We validated these proposed techniques on the Urdu sentiment analysis dataset and compare with Convolutional Neural Network (CNN), Recurrent Neural Network (RNN), and Long Short-Term Memory (LSTM). The results shows that the proposed technique LSTM-CNN has significantly by achiving testing accuracy 96.9% training accuracy 99.5, precision P 96.0 , precision N 98.0, Recall P 98.0, Recall N 96.0, F1-score N 97.0 , and F1-score N 97.0.The findings shows that CNN is a strong option for the given problem as it improves classification accuracy when combined with LSTM.

AI-Based Deepfake Audio Detection Technique from Real and Fake Audio Dataset

2025-03-21T14:36:33+00:00

The fast development of Deepfake technologies has created major challenges regarding audio authenticity throughout cybersecurity, along with journalism and individual privacy domains. Numerous studies investigate deepfake images and videos and the field of deepfake audio remains under investigation. Researchers need to study deepfake audio methods further because this field is not fully developed. The researchers attempted to develop an audio detection model for fakes despite encountering difficulties in this particular area. A deep learning-based framework was designed to conduct deepfake audio detection through the use of a Convolutional Neural Network and Long Short-Term Memory (CNN_LSTM) model which boosted detection efficiency. Our research included embedding and performing analysis through data preprocessing followed by classification of the Real and Fake datasets. This dataset is a combination of ASVspoof2021_LA_eval (Logical Access) and Deep-Voice datasets. The proposed model's performance evaluation included the use of confusion matrices and an accuracy graph. The model demonstrates efficient audio originality discrimination capabilities and reveals important audio characteristics suitable for effective classification systems. The detection rates of developed algorithms are analyzed through the evaluation of false positive and negative probabilities. The results provide an effective base for deepfake audio detection systems to achieve 97.0% accuracy in detecting false audio content. This development improves the authenticity assessment of audio materials when considering technological manipulation.

Effective Study of Design Perception of Learning Artifacts for Online-Learning Systems

2025-04-07T08:15:59+00:00

There are several different artifacts for designing a website and they play a very prominent role in the aesthetics and inducement of any website. So, the website must look attractive and appealing to the eyes of its users. This research work is basically about revealing the aspects that directly or indirectly affect the user’s cognitive beliefs, emotions, and feelings while using the website, whether the website’s attributes lead to positive emotions such as satisfaction, ease/convenience to use, and appealing or the negative emotions such as frustration, irritation, satisfaction, or annoyance. The actual focus is on web-based learning applications to make them less frustrating and more appealing, especially for the students as well as teachers. For that, two learning websites are taken “Coursera and Edx” and analyzed, the analysis is based on a survey after letting the participants use these websites to get their cognition towards these websites, and after that user’s perception/preference was conducted by finding the relationship between the attributes of both of the websites. Some artifacts that mainly tried to hit are aesthetic aspects (color, typography) and information architecture (content quality, navigation, interactivity).

Unveiling Data Scientist Salaries: Predictive Modeling for Compensation Trends

2025-03-04T06:03:11+00:00

The fast pace of artificial intelligence growth with big data has rendered data science as one of the most in-demand jobs in the world. Data scientists' remuneration structures, though, demonstrate significant heterogeneity by region, industry, and experience, thereby making career advancement difficult for both new and old entrants. Current studies tend to be based on small data samples or basic statistical techniques, hence neglecting the intricacies of determining the determinants of salaries. This study aims to utilize advanced machine learning techniques, including decision trees, ensemble techniques, and eXtreme Gradient Boosting (XGBoost), to build an inferential model of classifying and estimating data science salaries using important determinants such as experience, location, firm size, and job. The model suggested in this study achieves accuracy of 92.3% according to a Random Forest algorithm, which is higher compared to conventional regression-based techniques. Feature importance analysis reveals that experience accounts for 45.7% of salary variation, followed by firm size (22.8%) and location (18.6%). By being data-driven, this research gives practical suggestions to job seekers, organizations, and policymakers, hence allowing them to make informed workforce planning, salary negotiation, and talent acquisition decisions. The study contributes to the body of knowledge by improving the precision of salary classifications, determinants identification, and the usefulness of predictive analytics in labor market trend analysis.

IoT-Based System for Prediction of Fungal Disease Attack on Mango Leaves

2025-05-05T05:40:09+00:00

Mango has significant commercial importance in Pakistan, as it is the world’s sixth-largest mango supplier. However, mango is severely affected by fungal diseases due to variations in environmental parameters. These diseases are normally found in growing areas of mango worldwide; weather parameters such as humidity, temperature, and leaf wetness duration increase the probability of fungal infection on mango plants. Early prediction of fungal disease on mango leaves can help farmers prevent potential losses. The goal of the research was to use sensors based on Internet of Things (IoT) for the monitoring of mango orchards in real-time and collect environmental data from targeted areas for disease monitoring. The hardware was installed in the Mango Research Institute Multan (MRI) orchard to collect current data. The past data was collected from AWS-MNS University Multan. The study was then carried out by developing the logistic regression and random forest model for Anthracnose prediction using past and current weather data for predicting future Anthracnose infections. The accuracy of the logistic regression model was 96%, while the random forest achieved 99%. This study developed an IoT-based system to improve quality and quantity of mango production.

Enhanced Deep Learning Based X-Ray Analysis for COVID-19 Identification

2025-01-21T13:14:59+00:00

The rapid and accurate detection of COVID-19 is critical for mitigating its transmission and ensuring timely medical intervention. This research enhances COVID-19 detection by implementing the Artificial Neural Networks Algorithms. Our research paper embeds the concept of a Convolutional Neural Network for efficient and accurate detection of COVID-19 by taking x-rayed images of the lungs of patients as input. The proposed model development involves systematic steps, including data acquisition, preprocessing, and augmentation, as well as the application of a convolutional neural network to the prepared data. The dataset utilized for this research paper on COVID-19 detection using Artificial Neural Network (ANN) is obtained from the source of the website Kaggle. The dataset consists of three classes: normal, viral, and COVID-19 affected. The visual data of these three classes is utilized to train and test the model. PCR testing is the most used technique for COVID-19 detection, but this technique is pricey for people who belong to middle- or lower-class families, so our research paper overcomes this financial barrier by using X-ray images of the patient to detect whether the patient is infected with COVID-19 or not. Accurate identification of COVID-19 cases is vital for controlling its transmission. Minimizing false negatives ensures timely care for infected individuals, reducing spread. Our proposed model achieves an accuracy of 95% by using multiple layer Convolutional Neural Network.

IIOT: An Infusion of Embedded Systems, TinyML, and Federated Learning in Industrial IoT

2025-03-18T18:29:23+00:00

With the revolution of Industrial 5.0 the system was modified to smart manufacturing. This Industrial 5.0 is emerging with different technologies such as IoT which provide real-time monitoring, analysis, and data fetching. For the novelty in II0T Application, this article investigates the combination of embedded systems, Tiny Machine Learning (TinyML), and Federated Learning (FL). Data privacy is ensured by Federated Learning (FL), and local data processing becomes efficient through Tiny Machine Learning (TinyML). This infusion promises to decrease latency, increase productivity, and improve data security. As previously unsolvable issues or problems are being addressed with renewed enthusiasm, new paradigms for development and research are needed. The goal of this article is to provide a platform and overcome the knowledge gaps for future revolutionary research projects that will leverage the growing trends of embedded devices influenced by compressed artificial intelligence (AI) models [18]. Moreover, will discuss about the TinyML, federated learning (FL) that permits the models to be trained locally on edge devices by utilizing their data as well as reducing the requirement for centralized data accumulation that may be even impossible in fewer Internet of Things (IoT) situations. It charts the development of embedded devices and wireless communication technologies, demonstrating the advent of Internet of Things applications across a spectrum of industries. In addition, the paper conducts a thorough cutting-edge technology to find recent works that use TinyML models to readily available embedded devices, and talks about recent research trends.

Randomized Frame Selection Based Video Steganography Method for Secure Embedding of Secret Data

2025-04-28T09:26:08+00:00

Prior video steganography solutions are lacking in the randomized embedding process due to the known series of selection of video frames to divulge the secrecy of embedded data. Moreover, while maintaining a high Peak Signal-to-Noise Ratio (PSNR) is crucial for preserving video quality, many prior approaches fail to effectively balance security and visual integrity. This paper presents a novel video steganography method to introduce a randomized embedding process to achieve secure data hiding with enhanced PSNR. The proposed method utilizes secret key-based random frame selection and a least significant bit (LSB) embedding technique including encryption tactics to achieve effective and secure hiding of data within video files. A 64-bit secret key undergoes a series of various operations such as XOR, compliment, and logarithmic functions to derive a random frame number for data embedding. The plaintext message first undergoes encryption before being embedded through LSB substitution into the binary pixels of the selected frame. Experiments conducted on different video samples of varying dimensions demonstrate that the proposed method provides significantly improved PSNR [i.e. 74.15 dB] and lower mean squared error [i.e. 0.0002 dB] compared to previous techniques. This indicates enhanced imperceptibility, payload capacity and overall security of the embedded data. The proposed method addresses limitations in existing video steganography approachesrelated to static frame selection, data exposure, and insufficientevaluationmetrics. Withits robust encryption and high-fidelity data hiding, this technique has promising applications in military communications, access control systems, video archiving, and content authentication domains.

Advanced Next-Word Prediction: Leveraging Text Generation with LSTM Model

2024-12-11T21:05:13+00:00

Natural Language Processing (NLP) increasingly relies on machine learning to make better predictions of sequential text. This work focuses on the application of Long Short-Term Memory Networks, a variant of Recurrent Neural Networks that is specialized for modeling long-term dependencies. Traditional RNNs leave much to be desired in predicting sequences that contain repeated patterns or contextual dependencies. The research uses “The Adventures of Sherlock Holmes” as the training dataset and applies TensorFlow and Keras frameworks for implementation. The major preprocessing steps included word tokenization, n-gram creation, and one-hot encoding to prepare the dataset for modeling. The LSTM model was trained over 100 epochs to optimize prediction capabilities. Through this work, we show that LSTM is effective in next-word prediction and can potentially improve the performance and practicality of language models for real-world applications. The model achieved a commendable accuracy of 87.6%, demonstrating its effectiveness.

Optimized Skin Cancer Detection through Dermoscopic Imaging Using EfficientNetB4 Architecture

2025-03-19T18:15:27+00:00

Background: Skin cancer classification is a challenging task due to the fine-grained diversity in the appearance of various diagnostic categories. Detecting skin cancer at an early stage is vital for enhancing patient outcomes, as the prognosis for this condition greatly improves when diagnosed early. Convolutional neural networks have been found to be more effective than dermatologists in classifying multiclass skin cancer. Problem: The identification of skin cancer is frequently impeded by the subjective analysis of dermoscopic images, resulting in misdiagnoses and delayed treatments. The objective of this study is to create a reliable and effective classification system using the efficientnetb4 model, which will aid in early detection and ultimately enhance patient outcomes. Objective: The main goal of this study is to create a highly efficient and accurate classification system for skin cancer using the efficientnetb4 model. The goal of this system is to improve the accuracy of diagnoses, minimize misdiagnoses, and enable early detection of skin lesions, leading to better patient outcomes and a more efficient diagnostic process in dermatology. Methods: The EfficientNetB4 model is trained on the HAM10000 dataset using transfer learning and fine-tuning techniques on rotated images, zoomed in and out, and even flipped over to make variations. Then, it adjusted the hyperparameters in the fine-tuning step to fine-tune its weights so that the model could fit the classification task for skin lesions more precisely. Results: The leading model, EfficientNetB4, achieved a Top-1 Accuracy of 89.22%, a Top-2 accuracy of 88.82%, and a top-3 accuracy of 88.62%. Precision, recall, and F1 scores are computed for each class. This model has demonstrated excellent performance in melanoma (MEL) and benign kurtosis-like lesions (BKL). Criteria considering high-class imbalance were used in the assessment of Efficient Net classifiers. Models with an intermediate level of complexity, such as EfficientNetB4, demonstrated the most optimal performance. Confusion matrices were also discovered to be useful in identifying skin cancer varieties with the greatest capacity for generalization. Conclusion: Overall, EfficientNetB4 demonstrated superior performance in classifying multi-class skin cancer. Further development would be oversampling or synthetic data generation for even more class-balancing techniques to improve performance over underrepresented classes. More medical data, including images and clinical data, will probably increase the overall diagnostic accuracy.

Advance Assessment and Counting of Ripe Cherry Tomato’s Via Yolo Model

2025-01-10T10:26:30+00:00

The use of intensive labor measurement and the computer-based techniques for the gathering of phenotypic information in the laboratories has been in trend previously. This study focuses the detection and counting of the cherry fruit during the growth of the plant in the greenhouse. It's unique because it uses the deep learning method for the imaging of fruit instead of the classical computer techniques of imaging. The Yolo method imaging has been used to detect the different steps of the growth of the cherry fruit in the greenhouse rather than in the laboratory. This is an advance method which closely detects the object and each pixel of the object; so that the results of this study are comparatively better than the earlier works with the classical methods. The use of Yolo method in this study is rather an innovative step in the field of agriculture which will be helpful in future not only in the collection of phenotypic information, but it will also be useful in the automation of the processes such as harvesting. The results attained have successfully evaluated the detection of cherry tomatoes along with counting clearly. After obtaining the results have 92.6% precision rate and 94.7% recall rate in case of detection, counting and assessment (Ripeness and unripens). Overall study can help to manage the better yield and better quality product after valuable assessment of cherry tomatoes via algorithms.

AI-Driven Predictive Threat Detection and Cyber Risk Mitigation: A Survey

2025-02-07T10:46:41+00:00

Predictive analytics is revolutionizing cybersecurity and various industries by leveraging artificial intelligence (AI) and machine learning (ML) to enhance threat detection, risk mitigation, and decision-making processes. By enabling a shift from reactive to proactive security strategies, AI-driven predictive models improve the accuracy of cyber threat detection, reduce response times, and strengthen overall resilience against evolving attack vectors. Advanced techniques such as deep learning, anomaly detection, and natural language processing (NLP) enhance the adaptability and precision of these systems. A comprehensive review of existing research highlights key advancements, challenges including data integrity, algorithmic bias, and scalability and ethical concerns related to privacy, fairness, and transparency. Beyond cybersecurity, predictive analytics optimizes efficiency across sectors such as healthcare, finance, manufacturing, and energy, supporting smarter resource allocation and operational improvements. The integration of emerging technologies, including quantum computing, federated learning, and blockchain, further enhances predictive capabilities while ensuring security and compliance. By addressing these aspects, this research provides valuable insights to advance AI-driven predictive analytics, guiding the development of intelligent, ethical, and scalable solutions for a rapidly evolving digital landscape.

Intelligent Cyber Security Framework for Threat Detection using Ensemble Learning Techniques

2025-05-11T08:15:01+00:00

Cyber security is critical in today’s fast-paced digital landscape. As AI-driven solutions become indispensable for safeguarding enterprises, the escalating volume and complexity of cyber threats frequently overwhelm conventional security measures, resulting in significant financial and reputational risks. To address this challenge, this study proposes an advanced cyber security framework based on an ensemble learning model that combines machine learning and deep learning algorithms. Using the HIKARI-2021 dataset (Kaggle), we evaluated and compared multiple classifiers, including Random Forest, Decision Tree, Gaussian Naive Bayes, K-Nearest Neighbors, Logistic Regression, Multi-Layer Perceptron, and Convolutional Neural Network. By integrating these models through an ensemble approach, we leveraged their complementary strengths, achieving a notable 96.32% accuracy—a significant improvement over individual models. Beyond accuracy, the ensemble method enhances adaptability, enabling more dynamic and resilient security frameworks. Our findings highlight the efficacy of ensemble learning in cyber security, demonstrating its potential to fortify digital enterprises against evolving threats. This research not only advances practical solutions but also paves the way for future studies on AI-integrated cyber security, fostering innovation and robust digital infrastructure globally.

Incremental Learning with Self-Organizing Bayesian Adaptive Incremental Network (SOBAIN)

2025-02-17T08:26:27+00:00

Neural networks and artificial intelligence have revolutionized how machines learn by mimicking aspects of human cognition. One key area in this field is the ability of agents to understand and imitate actions involving various objects, allowing them to pick up new skills by observing others. Understanding and imitating how others interact with different objects has become a big topic since it allows learning new skills by simply watching others. By constantly updating their own knowledge, lifelong learners can build on what they know over time. But the environments that artificial agents deal with are very different. Existing models designed for “lifelong learning” typically work with simplified experiments and datasets made up of static images, which limits their effectiveness for real-world applications. In this study, we propose a developmental model focused on how agents can learn about objects and actions through sensorimotor feedback, enabling humanoid robots to mimic actions more naturally. Our approach, "SOBAIN," works in three stages: neuron activation, neuron matching, and neuron learning. First, neurons activate based on specific traits that determine if they should "fire" or not. Then, we match the best neuron by comparing activation levels. At each learning point, new neurons connect to the network to match learned data. The final stage uses this network to refine and apply learned information, helping address memory loss issues in lifelong learning. By using these techniques, SOBAIN helps robots adapt their upper body movements in line with the sensor's feedback, creating a chain of neurons that builds over time as the robot learns new actions.

Textile Defect Detection in the textile industry using Deep Learning

2025-04-19T18:50:06+00:00

In order to maintain high production standards, the quality control process in textile manufacturing mostly depends on the efficient detection of fabric flaws. Conventional defect monitoring techniques are labor-intensive, manual and prone to human mistake, which results in inconsistent quality. In this research, the hybrid deep learning model is proposed using convolutional neural networks and gated recurrent unit networks for textile defect detection in the textile industry. The goal is to increase this crucial process's precision, effectiveness, and dependability. Since fabric defect identification is a crucial step in quality control, it is one of the manual operations that has gradually been automated using the aforementioned techniques. The performance evaluations were conducted on proposed model and compare with other models including Convolutional Neural Networks (CNNs), Artificial Neural Networks (ANN), Recurrent Neural Networks (RNN), and Long Short-Term Memory (LSTM). CNN ability to extract features and GRU skill at sequential learning credited with this improved performance, which allow the model to successfully capture temporal and spatial relationships. The proposed hybrid model which combines CNN + Gated Recurrent Unit (GRU) are performed well as compare to other models and achieved the accuracy of 0.9841. The findings shows that CNN is a strong option for the given problem as it improves classification accuracy when combined with GRU.

A Survey: Enhancing Wireless Security in the Digital Age

2025-02-13T12:05:38+00:00

In today’s globalized society, protecting wireless networks from unauthorized users is fundamental. Wireless networks and mobile devices are under constant attack, far more than their wired equivalents, which has made achieving complete protection an unfulfilled goal, despite decades of research. This study considers critical matters such as authentication, confidentiality, integrity, and availability, which are important in the security of wireless networks. We study existing protocols and recommend changes to improve their security. An innovation consists of a new strategic approach to security risk management integrating a data authentication and integration model with machine learning methods. This also provides an examination of the current responses to wireless network security problems, highlighting their advantages and drawbacks in light of constant changes to cyber threats, especially phishing attacks. The objective is to accentuate the need to strengthen wireless networks security to safeguard confidential information from unauthorized access. Additionally, the article provides an in-depth review of methods to address security vulnerabilities associated with wireless networks.

Citrus Fruit Postharvest Losses Analysis using Hybrid Features

2025-04-29T18:55:36+00:00

The citrus fruit is an essential food source for humans, as it contains several vitamins and minerals, which makes it one of the most significant fruit crops globally. A wide variety of fascinating citrus fruits can be found in Pakistan. Most postharvest losses, 30-50% of the total production, are due to improper handling of fruits throughout the packing, shipping, and storing processes. Manual sorting of horticulture items increases postharvest losses however innovative algorithms and technology offer solutions to mitigate these losses. The focus of this research is to enhance current methods of fruit grading, decrease post-harvest losses and preserve fruit quality. Artificial intelligence (AI) model grades citrus fruits for real-world applications. With the use of AI classifiers, we can categories citrus fruits into four distinct stages based on their ripeness: initial-ripe, semi-ripe, fully ripe, and defective or damaged. Each stage is distinguished by its own unique hybrid feature, color attributes (RGB), texture and size. Three Convolutional neural network (CNN) models namely, MobileNet, VGG16, and Inception V3 used to achieve accuracy of 97%, 98%, 95.6% on citrus fruit dataset. The VGG16 gives the best results on grading.

Advanced Collaborative Robotics: Enhancing Industrial Cobots with Conversational Interaction and Computer Vision

2025-03-24T10:51:22+00:00

Industry 4.0 has revolutionized modern manufacturing by integrating advanced automation, data exchange, and smart technologies. At the forefront of this transformation are collaborative robots (cobots), which have become indispensable components of smart manufacturing systems due to their flexibility and safety features. This study introduces a novel approach that enhances cobot functionality by merging conversational AI and computer vision, enabling adaptive human-robot interaction and dynamic task execution in industrial environments. The proposed system leverages a transformer-based conversational module designed to facilitate natural language communication between human operators and cobots. This module allows workers to issue commands, receive feedback, and seamlessly coordinate complex tasks without requiring specialized programming skills. In parallel, a YOLOv5-based vision module is integrated for real-time object detection and defect identification, significantly improving situational awareness and task precision. Comprehensive evaluations demonstrate that the integrated system achieves a 33% reduction in task completion time compared to conventional cobot setups. Additionally, the object detection model attains a precision of 96.2% and a recall of 93.5%, ensuring reliable and accurate identification of objects and potential defects. These advancements significantly enhance task efficiency, accuracy, and overall usability, surpassing the current state of the art. Furthermore, the system's conversational capabilities empower operators to adjust processes dynamically, minimizing downtime and improving workflow management. The combination of conversational interaction and computer vision not only augments operational performance but also fosters a more intuitive and human-centric collaborative environment. This research highlights the transformative potential of integrating conversational AI and computer vision into cobots, paving the way for next-generation collaborative robotics within Industry 4.0 applications. However, challenges remain in optimizing contextual understanding and refining vision algorithms to further boost performance and adaptability. Future work will focus on enhancing the cobot's ability to process complex contextual cues and improving the robustness of object recognition under variable conditions. By addressing these challenges, the proposed system will better meet the demands of dynamic industrial settings and continue to shape the evolution of human-robot collaboration.

Jellyfish Algorithm for Feature Selection: Improving Machine Learning-Based Heart Disease Prediction

2025-03-20T14:51:18+00:00

This research paper describes a sophisticated medical diagnosis system based on machine learning (ML) to predict heart disease. The Jellyfish algorithm optimizes the Cleveland dataset, which aims to achieve the most accurate predictions with the most significant features chosen. The selection of features is crucial for performance as there can be excessive features causing overfitting or too few features causing accuracy loss. The jellyfish technique is type of swarm metaheuristic approach in which the feature selection is optimized and performance of the model is enhanced. After selecting features, we will train four machine learning algorithms on the optimized dataset and program. The algorithms are Artificial Neural Networks (ANN), Decision Tree (DT), AdaBoost, and Support Vector Machines (SVM) Results show that every single model benefits from feature selection. The feature selection mostly impacts the Support Vector Machine model which sees the highest increase from 98.09% to 98.47%. Every performance metric has a respective enhancement in Sensitivity, Specificity and Area under the Curve (AUC) which shows that Jellyfish can enhance the accuracy of heart diseases prediction.