An Enhanced AI Framework for Predicting Type II Diabetes – Insights from 100000 Health Records

Abstract

Diagnosis of Type 2 diabetes at an early stage is very essential towards effective prevention and treatment of the disease particularly among high-risk groups. This study leverages the state of the art Machine Learning (ML) models and Deep Learning (DL) models to come up with an effective and accurate diabetes risk prediction framework based on the Diabetes Risk Prediction dataset available on Kaggle, which contains 100000 samples with a rich combination of health indicators. Eight types of classifiers i.e. Logistic Regression (LR), Support Vector Machine (SVM), K-Nearest Neighbors (KNN), Random Forest (RF), XGBoost, LightGBM, CatBoost, and TabNet were evaluated used and the results were compared. The executed models were evaluated on various key performance metrics. TabNet model beat the rest of the models (91.84% accuracy and 0.962 AU-ROC) closely followed by CatBoost (91.66% accuracy) and XGBoost (91.54% accuracy). LightGBM is a little bit less accurate (91.10%), but showed relatively good results and improved compared to traditional solutions, including Random Forest (91.08%) and Logistic Regression (78.88%). These findings prove the high predictive accuracy of TabNet, has great potential as a healthcare solution in the real world in the case of Type 2 diabetes detection.

Ähnliche Arbeiten