Enhancing Diabetes Mellitus Prediction: Integrating Hybrid Deep Learning Model with Sampling Techniques

Diabetes, characterized by high blood glucose levels, is a leading cause of liver, eye, kidney, and heart diseases. This study evaluates various deep learning models, combined with machine learning classifiers, for predicting diabetes mellitus using the BRFSS dataset. The dataset's imbalance posed a challenge for binary classification, common in medical diagnostics. To address this, different sampling techniques were tested. Hybrid models combining Convolu-tional Long Short Term Memory (Conv LSTM) networks with traditional classi-fiers were also explored. The Conv LSTM model combined with Adaboost classifiers achieved the highest accuracy of 89.47% with SMOTE-ENN resampled data. These findings highlight the potential of integrating deep learning and traditional machine learning for effective diabetes prediction, aiding early diagnosis and intervention