Enhancing Diabetes Prediction using Hybrid Feature Selection and Ensemble Learning with AdaBoost

Abstract

This research relies on a feature selection method termed LasNO to analyse the progress of diabetes prediction. The ANOVA integrated with Lasso regularization identifies significant predictors like Age, BMI, DiabetesPedigreeFunction, and (Glucose). An AdaBoost model with the following characteristics was developed, and the results were exceptional performance metrics: 97.02% accuracy, 87.65% precision, and 82.5% recall. The results of the primary focus on glucose levels to differentiate between events with and without diabetes have been affirmed from the visual analyses. It also calls into attention that due to the appropriate cross-validation, LasNO is very successful on improving the interpretability and the possibility of generalization of the model. Such findings make it clear that efforts towards early detection of the condition and subsequent timely implementation of measures to contain the illness can go quite a long way in enhancing the quality of the lives of people that contract the disease. Further research is likely to be devoted to the assessing of the improved and broadened validation of the dataset as a critical approach for the enhancement and expansion of the model for use in clinics. As a result, this study aims to advance diabetes theragnostic possibilities and healthcare recommendations for better action based on ML approaches enhancing the accuracy of predictions. In this study, approaches to feature selection used for improving the performance while also making the models more understandable are the focus. The idea is to ensure that medical practitioners have reliable tools to undertake early detection and management of diabetes. Thus, the main objective of dealing with patients’ Health-Related Quality of Life is to simultaneously increase the quality of patients’ living while decreasing the costs of their treatment.

Topics & Concepts