Analysis of Feature Extraction Algorithms for Artificial Intelligence Based Diabetes Detection

Diabetes is among the fastest-growing health issues worldwide, hence necessitating proper predictive models. This research focuses on the development of an AI-based predictive model for diagnosing diabetes based on the Diabetes 130-US hospitals dataset. Various preprocessing techniques are conducted, including data cleaning, encoding, and imputation of missing values. To balance the classes in the data set, the Synthetic Minority Over-sampling Technique (SMOTE) is employed. The feature extraction process is done by applying the Random Forest technique, while the selected features are classified through the XGBoost classifier. Model performance is analyzed with the help of accuracy, precision, recall, and F1 score metrics. As per experiment findings, performance of the model gets improved due to feature extraction. The value of accuracy is raised from 78% to 79% due to feature extraction. Likewise, there is improvement in the case of precision, recall, and F1 score as well. Proposed methodology makes prediction more reliable with less computational cost. It will be helpful for healthcare professionals in diabetes detection at an early stage.