Optimization of Diabetics Diagnosis Using SVM, Random Forest and XGBoost

Diabetes is one of the common chronic conditions in today's world. Early and accurate diagnosis of diabetes helps in its effective management and prevention of complications. This paper aims at finding optimized machine learning model for diabetic prediction. Three machine learning models, Random Forest, eXtreme Gradient Boosting (XGBoost) and Support Vector Machine (SVM) are considered to evaluate the predictive performance on the diabetic dataset considered from UCI machine learning repository consisting of health-related parameters for identifying whether an individual is diabetic or not. The assessment parameters considered for evaluation are Precision, Recall, Accuracy and F1-Score. The algorithms are run on training datasets of varied sizes, the results are analyzed through performance comparison among three models. The findings of this work based on the performance analysis are: SVM performed better for large datasets and handled imbalanced and high-dimensional data well therefore it is appropriate for hospital databases. Random Forest displayed consistent performance irrespective of the dataset size, giving robustness with interpretability from feature importance analysis and hence can be used to improve clinical decision support systems. XGBoost was very efficient for smaller datasets because of its gradient-boosting mechanism that optimizes predictions efficiently and therefore it is very effective for real-time use like wearable health monitoring. The results are also compared with advanced deep learning methods and it is observed that machine learning models make better option on real-time datasets due to the ease of implementation, speed, and resource efficiency.