Ensemble Methods of Machine Learning Algorithms for Early Diabetic Detection in Comparison

Diabetes can be an issue for people of all ages. Even everyone people among the nine is diabetic positive, which can bring more complications to humans. Furthermore, people in Bangladesh are hesitant to talk about this illness and frequently need clarification. They don't care until their life is in danger. Therefore, we researched diabetes prediction through machine learning so that people can know if they are diabetes positive quickly. The main goal is to contribute to the development of an automated system for diabetic prediction which can predict early diabetes using some basic human parameters. One of the significant parts of Machine Learning is Feature Selection techniques. In this work, we intend to focus on this part of the ML paradigm. We analysed the research data and applied feature selection techniques to enhance the performance of ML algorithms. We use PCA, Chi-square test and F_classif as our feature selection technique. Then we use the LGBM Classifier, XGB Classifier, SVC, Random Forest Classifier, and Gradient Boosting Classifier and determine the accuracy to justify our selected techniques. Moreover, We applied a voting classifier and weighted average on these models. Determining the weighted average of ensemble ML methods is a new approach to diabetic prediction. The weighted average outperformed other models with an accuracy of 96.10%. However, the main goal of this research is to provide a comparison scenario of ML methods in diabetic prediction.