An Improved Machine Learning Model for Early Detection of Vein Thrombosis

Abstract

Deep Venous Thrombosis (DVT) is a vascular disorder requires early and accurate prediction to prevent serious complications such as pulmonary embolism. Although medical diagnostics have advanced significantly, existing predictive models still struggle with issues such as data imbalance and limited generalization, resulting in challenges for developing scalable and reliable prediction systems. This research aims to address these limitations by suggesting a hybrid machine learning model that integrates clustering, sampling, and ensemble classification techniques to enhance DVT prediction accuracy. The experimental design employs Agglomerative Hierarchical Clustering, the Synthetic Minority Over-sampling Technique (SMOTE), and a Stacking Ensemble classifier composed of Decision Tree (DT), Stochastic Gradient Descent (SGD), Quadratic Discriminant Analysis (QDA), and Naive Bayes (NB) as base learners, with Logistic Regression serving as the meta-learner. Model was evaluated using accuracy scores and confusion matrices to assess classification reliability and error rates. The proposed hybrid model achieved an accuracy of 97.68%, with the lowest false-negative rate, confirming the diagnostic effectiveness of the DT-based hybrid approach. The results demonstrate that algorithmic integration can significantly enhance predictive accuracy and robustness in clinical applications. Overall, this hybrid framework bridges both practical and scientific gaps by offering a scalable and interpretable solution for the initial level detection of DVT. Future work will focus on incorporating temporal data and validating the model in real-world clinical environments.