Pre-Diabetic Diagnosis from Habitual and Medical Features using Ensemble Classification

Abstract

Diabetes is the fastest growing metabolic disorder. It has become a serious global health emergency. This disease if not diagnosed can increase the risk of serious life-threatening diseases like cardiac arrest, brain bleeds also known as brain strokes, kidney damage and many more. In this paper, we considered a challenging problem of” Pre-Diabetes Detection” using different deep learning and machine learning algorithms. The prime target of our study is to detect diabetes at an early age so that we can help the patient to properly diagnose this disease and better take care of their health. This is important because if we can correctly predict and then provide diagnosis of Prediabetes there is a 58% chance that the person can go back to his healthy life. So, considering the power of deep learning and machine learning we intend to exploit/explore parameters/features-space using suitable classification techniques. For this we employed several techniques i.e., LSTMs, MLP and ANN for deep learning and also Random Forest (RF), Logistic Regression (LR), Decision Tree (DT), KNN and Gaussian Naive Bayes (GNB) for machine learning respectively. We also performed majority voting from all Machine Learning and deep Learning models and made an ensemble model from them to evaluate performance. Among these techniques our recommended ensemble model for deep learning and LSTMs outperformed because they were very effective than other standard machine learning and deep learning model. To validate the proposed technique, we used a standard publicly available dataset. We have then applied majority voting and made an ensemble classifier from that and evaluated the results.