Different machine learning methods based prediction of mild cognitive impairment

Abstract

Aim: In this study benefits from different machine learning methods to analyze factors which affect young person’s scores ofcognitive assessment.Material and Methods: This study was performed among 144 persons aged between 18 and 24 who study at Kahramanmaras SutcuImam University. Boosted Tree Regression (BTR), Random Forest Regression (RFR) and Support Vector Machine (SVM), which areamong machine learning methods, were used in order to determine the factors affecting the score of cognitive assessment. K-10fold cross validation method was also used. Mean Squared Error (MSE), Root Mean Squared Error (RMSE), Mean Absolute Error(MAE) and Correlation coefficients (R) metrics were used in order to measure prediction performances of machine learning methods.Results: MSE values were calculated as 9.66 for BTR, 9.78 for RFR, and 6.43 for SVM. MAE values were calculated as 2.06 for BTR,2.05 for RFR, and 1.97 for SVM. RMSE values were calculated as 3.10 for BTR, 3.12 for RFR, and 2.53 for SVM. Finally, correlationcoefficients were calculated as 0.289 for BTR, 0.371 for RFR and 0.546 for SVM. In addition, it was also found out that the mostimportant variables which affected the scores of cognitive assessment were anti-depressant use, depression and obsession.Conclusion: It was demonstrated in this study that SVM displayed the lowest error rates and highest prediction performance in termsof determining the score of cognitive assessment. Therefore, SVM can be stated that it is the most suitable method for the predictionof cognitive impairment.