Comparing Supervised and Unsupervised Machine Learning Techniques to Decision Support Systems in Healthcare

This article assesses the effectiveness of various machine learning algorithms in decision support systems within the healthcare sector. The focus lies on predicting the diagnosis of type II diabetes using multiple machine learning techniques. The findings indicate that supervised techniques, particularly Random Forest, outperformed unsupervised techniques in terms of accuracy and overall performance. The implementation of Ensemble Learning in the Random Forest algorithm yielded superior results when compared to Logistic Regression and SVM Linear. The latter two algorithms showcased diminished performance due to their reliance on linearity assumptions. Naive Bayes also demonstrated lower performance, primarily attributed to its assumption of mutual independence between features. Unsupervised techniques, which operate on unlabelled data, exhibited inferior performance compared to their supervised counterparts. Additionally, the application of unsupervised clustering preprocessing techniques did not lead to improved performance, although it did have a positive impact on overall results. In conclusion, this study highlights the superiority of supervised techniques over unsupervised techniques in the context of predicting type II diabetes. Furthermore, it emphasizes the significance of considering false negatives when evaluating the performance of machine learning algorithms.