A Comparative Analysis of the Performance of Parallel Ensemble and Sequential Ensemble Machine Learning Methods in the Detection of Diabetes Miletus

Diabetes Mellitus still forms a major cause of death rates soaring around the globe, heightening scares regarding shooting up diabetic population in the world; and hence straining health attendants to seek for rapid diagnostic tools specific to an incurable disease as described. Many models have been presented for machine learning as base learners, or else combined ensemble techniques. The performance of parallel and sequential ensemble machine learning approaches in the detection of diabetes mellitus: A comparative study, the parallel ensemble methods include Random Forest, J48, CART and Decision Stump (DS) classifiers and the sequential ensemble method includes XGBoost AdaBoostM1 Gradient Boosting. The data set was 70% training and 30 % testing using the dataset on UCI machine repository site. Python analysis using Jupyter Notebook of this model confirmed that sequential ensemble has a classification accuracy about 6% more than parallel method using the same dataset by applying the 5-fold Cross Validation (CV) technique. XGBoost was also 4% better than 10-fold CV. Sequential machine learning models perform better in predicting diabetes mellitus as per the results. Therefore, the study concludes that sequential ensemble approaches are robust and effective in enhancing early diagnosis of patients. Thus, these models can be employed to develop prospective diabetes mellitus detection systems which in turn contributes to better health outcomes and decreasing the load on healthcare.