Semi-Supervised Learning Approach Using Hybrid Bidirectional LSTM Networks for Diabetes Prediction with Explainable AI

Diabetes is a serious health condition that can result in serious complications if not properly detected or treated. Despite recent developments in machine learning and deep learning, which have enhanced the detection of diabetes, there are still challenges. It is necessary to overcome challenges such as data imbalance and enhance key performance measures for the creation of reliable predictive models. In the present study, we suggest a semi-supervised learning strategy with the pseudo-labeling technique and a hybrid Bidirectional Long Short-Term Memory-Gated Recurrent Unit model that is tailor-made to overcome data imbalance and optimize overall prediction performance. Additionally, in order to achieve an additional boost in the model’s performance, Random Under-Sampling (RUS) is employed, leading to a significant improvement in the F1 score as well as AUC. Our model achieves an AUC of 0.8297, accuracy of 0.7503, and F1 score of 0.7649. Additionally, we cover how to use explainable AI using SHAP (Shapley Additive Explanations) for interpretation of feature contributions, providing greater transparency in the decision-making of the model.