SMART WEB APPLICATION FOR DIABETES PREDICTION USING MACHINE LEARNING AND DATA STORAGE

ABSTRACT

Diabetes is a common chronic illness, and controlling complications and enhancing health outcomes depend heavily on early detection. Through the use of machine learning, intelligent solutions for diabetes diagnosis and prediction can be developed. To obtain a high predicting accuracy, the project developed a smart web application that combines machine learning models with a secure data storage architecture. KNN, Random Forest, SVC, Decision Tree, and XGBoost are the five supervised machine learning algorithms that were used. The primary issue that needed to be addressed was the dataset's imbalance, which included more patient records that were not diabetic than those that were, potentially distorting the model's predictions. MinMaxScaler feature scaling and SMOTE oversampling were applied to the 2000 patient records to balance the dataset. Feature correlation analysis was utilized to identify diabetes predictors that were of importance. After the algorithms were trained and improved, the model created using the Random Forest algorithm showed the highest level of reliability with a 98.5% accuracy and 98.5% F1-score. For real-time diabetes prediction, the trained Random Forest model was included into a Streamlit web application, which lets users enter data safely in a MySQL database. This smart program integrates machine learning, user-friendly interfaces, and data storage for early diabetes detection, better healthcare decision-making by resolving dataset imbalance.