Comprehensive Analysis of Diabetes Prediction using Machine Learning and Deep Learning Approach

Diabetes is a widespread disease characterized by abnormal blood glucose levels, which significantly reduces the quality of human lives and can lead to death because of its complications. Diabetes prediction and treatment are important to reduce and avoid associated risks. This paper presents a survey of existing Machine Learning (ML) and Deep Learning (DL) frameworks for diabetes prediction, focusing on their techniques and performance. Two publicly available datasets, namely, the PIMA Indian Diabetes dataset and the Early-Stage diabetes risk prediction dataset containing clinical attributes such as blood glucose levels, are utilized for training and testing models. A ML-based method such as Bayesian Network and KNearest Neighbor (KNN) are employed for feature grouping and distance-based learnings. A DL-based methods such as lightweight Artificial Neural Network (ANN), autoencoder, and attention-based mechanisms are utilized to capture complex features. A comparative analysis provides the advantages, limitations, and performance metrics of existing methods. The performance is evaluated based on the performance metrics such as accuracy, precision, recall, and F1-score.