One of the most common chronic metabolic diseases all over the world is Diabetes mellitus and also this disease may cause complications, such as heart disease, renal damage, and visual impairment. Therefore, the precise and earliest diagnosis of diabetes is extremely necessary to prevent or
minimize such complexities. In this study, a smart machine learning model is proposed for the earlier detection of diabetes mellitus based on the Pima Indians diabetes dataset, which includes health information of diabetic patients (glucose level, blood pressure, skin thickness, insulin level, body mass index (BMI), diabetes pedigree function, pregnancies and age). This work includes Four sections namely Exploring Data to explore the datasets and analyse it. Feature selection/extraction in which, after feature selection, the dataset features will be Normalized and a SMOTE with ENN to reduce class imbalance with proper selection criteria to enhance accurate prediction of samples. Selected and trained different classifier to predict the results in four MLs, i.e., Logistic Regression, Support Vector Machine, Decision tree and Random Forest, also calculate feature importance on the respective models to show highest impact of features. We proposed a simple and efficient framework that combines data pre-processing technique(s), and the comparative prediction performed on different MLs models with feature analysis to make an appropriate prediction of diabetes. Experiments result showed that Random Forest provided best outcomes with highest accuracy of 95.52% , best precision of 96% , best recall of 95% and best F1-Score of 96%, out of rest classifiers..
We determinedglucose, BMI, age, and insulinas the top significant features for early prediction of diabetes. The proposed framework can act as useful and ready to deploy tool as the support system for predicting Diabetes mellitus and its further personalized diagnosis for the patients.