Optimizing Diabetes Prediction Using Machine Learning with Data Deviation

Diabetes mellitus is a chronic disease that affects global health with an increasing prevalence rate. Early prediction can provide opportunities for earlier intervention and better management of health conditions. This study explores various classification techniques to analyze data sets that include potential risk factors such as family history, body mass index (BMI), blood pressure, and blood sugar levels. This approach uses several machine learning algorithms such as Decision Trees, Gradient Boosting, support vector machines (SVM), K-Nearest Neighbor (KNN), Naive Bayes (NB) and Random Forest (RF) to identify patterns related to the development of diabetes. In this study, we have carried out a comparison of several machine learning (ML) algorithms, aiming to classify the “No Diabetes” or “Diabetes” class by considering several parameters such as, Age, gender, polyuria, polydipsia, sudden_weight_reduction, weakness, polyphagia, genital_thrush., visual_blurry, itching, irritability, delayed_healing, partial_paresis, muscle_stiffness, alopecia, obesity and classes collected in one dataset. The ML model is optimized by normalizing standard deviation data and 20-fold cross validation to get the best model and to determine the log loss or accuracy matrix and its performance. The results show that the model built using Gradient Boosting and SVM gave the best accuracy results on the entire test set, the Decision Trees, Random Forest and k-NN models also gave good results, while Naïve Bayes proved to be the least good in terms of accuracy results.