Evaluation of Machine Learning Techniques for Classifying and Balancing Data on an Unbalanced Mini-Mental State Examination Test Data Collection Applied in Chile

The Mini-Mental State Examination (MMSE) is the most widely used cognitive test for assessing whether suspected symptoms align with cognitive impairment or dementia. The results of this test are meaningful for clinicians but exhibit highly unbalanced distributions in studies and analyses regarding the classification of patients with cognitive impairment. This is a complex problem when a large number of MMSE tests are analysed. Therefore, data balancing and classification techniques are crucial to support decision-making in distinguishing patients with cognitive impairment in an effective and efficient manner. This study explores machine learning techniques for data balancing and classification using a real unbalanced dataset consisting of MMSE test responses collected from 103 elderly patients participating in a Chilean patient monitoring project. We used 8 data classification techniques and five data balancing techniques. We evaluated the performance of the techniques using the following metrics: sensitivity, specificity, F1-score, likelihood ratio (LR+ and LR-), diagnostic odds ratio (DOR), and the area under the ROC curve (AUC). From the set of data balancing and classification techniques used in this study, the results indicate that synthetic minority oversampling and random forest balancing techniques improve the accuracy of cognitive impairment diagnosis. The results obtained in this study support clinical decision-making regarding early classification or exclusion of older adult patients with suspected cognitive impairment.