Diabetes Prediction Using Machine Learning Algorithm: A Comparative Analysis

Diabetes prediction is an essential task in healthcare that could be achieved through Machine Learning models. Several factors contribute to diabetes such as overweight, high cholesterol levels or frequent urination. This research includes variety of data sources for predicting diabetes including The RTML dataset consisting data from Bangladeshi women and PIMA Indian dataset. These both combined thereby enriching the dataset's variety and data sources. The mutual information feature selection approach was integrated into this research to find the most relevant features in the dataset. To tackle the issue of class imbalance over-sampling Technique (SMOTE) is utilized to rectify the uneven distribution between diabetic and non-diabetic populations. The classification accuracy of the different Machine Learning(ML) algorithms is compared which include algorithms like random forest, logistic regression, voting classifier, KNN, neural networks, Ada Boost, XgBoost and voting classifier. After testing all the algorithms, the model achieved a cross validation score of 85%, accuracy of 83.22% and AUC of 0.86 using the XgBoost algorithm. The use of explainable AI techniques like the LIME framework helped provide interpretable explanations of the model's predictions.