Enhancing a System for Predicting Diabetes Utilizing Conventional Machine Learning Approaches

Diabetes is a prevalent sickness that involves millions of people worldwide. Despite several attempts to develop a precise model for predicting diabetes, there are still substantial unresolved research challenges. These challenges arise from the absence of suitable datasets and effective prediction methods. This has led scientists to employ machine learning-based methods to conquer these problems. This work aimed to explore how analyzing features and algorithms of machine learning might be employed in diabetes by applying five different machine-learning methods. The process of classifying may encounter obstacles due to certain features that may lead to difficulties. To overcome this issue, Logistic Regression (LR) has been utilized to determine the degree of effect of these attributes on predicting diabetes mellitus. In order to classify, the proposed diabetes prediction model applies various classifiers such as Adaptive Boosting (AdaBoost), k-nearest neighbor (KNN), linear discriminant analysis (LDA), NaïveBayes (NB), and Support vector machine (SVM) to the Pima Indian Diabetes (PID) database. Furthermore, the performance of each algorithm is scrutinized to determine the one with the highest accuracy, specificity, and recall. It was discovered that the NB model is more effective for binary classification with a refined selection of attributes, whereas random forest is more proficient in handling more features.