A Machine Learning Approach Using Feature Selection and Scaling with Hyper Parameter Tuning Method for Early Prediction of Diabetes

Introduction: Diabetes mellitus is an enduring condition characterized by raised blood glucose levels and has become a significant global health concern. Early and accurate diagnosis is crucial and that can help to prevent or delay of problems like cardiovascular diseases, kidney complications, nerve impairment, and vision diminishing due to diabetes. Diabetes is a chronic metabolic disorder that affects millions of people worldwide, leading to severe health complications if not diagnosed and managed in its early stages. Early prediction of diabetes is crucial for timely intervention and personalized treatment, reducing the risk of long-term complications. Traditional diagnostic approaches rely on clinical tests, which may not always be efficient in identifying high-risk individuals before the onset of the disease. With advancements in artificial intelligence (AI) and machine learning (ML), predictive models have gained prominence in healthcare applications, offering improved accuracy and efficiency in disease diagnosis. However, the performance of these models heavily depends on the quality of input features, data preprocessing techniques, and hyperparameter tuning strategies. Objectives: The main objective of this Research work is to predict diabetes at an early stage so that any severe complications may avoid. Methods: The most significant and robust features of the dataset are chosen using the attribute selection tool and correlation attribute estimation method by using the WEKA software tool. Then, features form the dataset are scaled using the standardization feature scaling technique and different ML classification algorithms such as LR, KNN, Naïve Bayes, Support Vector Machine, Decision Tree and Random Forest are used for experimenting with the above machine algorithm on the PIMA Indian diabetes dataset in PythonIn the preprocessing method, identification and removal of null and duplicate values have been replaced with the mean values. Results: By applying six different machine learning algorithms on PIMA Indian diabetes Dataset have shown that the K-Nearest Neighbor and Naïve Bayes both classifiers reported the maximum prediction accuracy of 81.82%, followed by LR, SVM, and RF with accuracies of 79.87%, 79.22%, and 77.27% respectively. Conclusions: By appropriate feature Selection and by hyperparameter tunning increase the diabetes prediction accuracy at an early stage.