Experiment 2

Although the accuracy for both machine learning models before resampling has scored better compared to the one after resampling, but the models after resampling scored better in precision, recall and f1-score.

This means that the models can predict more Normal MMSE compared to the previous one but not all are predicted correctly. This is acceptable because Normal MMSE is the minority among all 3 MMSE classes, causing the models difficult to learn the pattern of the Normal MMSE.

To compare the machine learning models after resampling, we think that the Random Forest model under SelectKBest feature selection technique is the best model we could have. This is due to its accuracy, which is the second highest among 4 of them, and also the total predicted Normal MMSE is higher than others but at the same time maintains the second highest scores for precision, recall and f1-score.