An Efficient Approach for Diabetes Prediction Through Integrated Feature Engineering and Machine Learning

Diabetes, a pervasive chronic health condition, affects millions worldwide, emphasizing the significance of prompt detection and efficient management to mitigate associated complications. Within this research, we suggest an efficient approach for early diabetes detection using clinical features. Our research explores the efficiency of several models for machine learning applied to the Indian PIMA dataset. Additionally, we integrate outlier rejection, missing data handling, and data balancing techniques into our approach to enhance its robustness and accuracy. Through rigorous experimentation and analysis, we evaluate the performance of models such as Gradient Boosting (GB), Logistic Regression (LR), Decision Tree Classifier, and Extreme Gradient Boosting (XGB). Our findings highlight the superiority of ensemble methods, particularly XGB, which achieved remarkable f1 score, recall, accuracy, and precision of 92%. Furthermore, we present a comparative summary with previous research on the same dataset, showcasing the advancements made by our approach. The results underscore the importance of data management and ensemble techniques in improving predictive accuracy.