Embedded Files
To find out which feature is most useful for prediction of MMSE and GDS-15, we conducted 15 experiments with different combinations of features that will be used to train the model
Exp 1: demographic
Exp 2: health
Exp 3: social
Exp 4: psychology
Exp 5: demographic + health
Exp 6: demographic + social
Exp 7: social + health
Exp 8: social + psychology
Exp 9: health + psychology
Exp 10: demographic + psychology
Exp 11: demographic + health + social
Exp 12: demographic + health + psychology
Exp 13: demographic + psychology + social
Exp 14: psychology + health + social
Exp 15: demographic + health + social + psychology
Next, we want to find out which model performs the best among the models chosen, and if different feature combinations will bring out different accuracy.
5 models used are such as Decision Tree, MLP classifier, KNN classifier, SVC classifier, Naive Bayes
PART A: Extract Top 10 of best features from each domain
For the prediction of MMSE, we found out that the model Decision Tree has the highest average accuracy among the 5 models and only selecting demographic features has the highest average accuracy among the 15 combinations.
The highest accuracy achieved is 70.06% with features of experiment 5 (demographic + health) using Decision Tree.
Thus it is safe for us to conclude that demographic features can affect the result of MMSE the most.
For the prediction of GDS-15, we can see that using various combinations and models can easily get high accuracy, except for the Naive Bayes model.
The MLP classifier model has the highest accuracy of 89.90% followed by the Decision Tree model with an accuracy of 89.17%, with only a slight difference of 0.73%.
Experiment 3 (Only selecting social features) gives the highest average accuracy among the 15 combinations of features.
The highest accuracy of 90.24% comes from the model Decision Tree and Experiment 4 (only selecting psychology features)
Thus, we can conclude that using the Decision model can give results with high accuracy to both the prediction of MMSE and GDS-15.
Either selecting experiment Experiment 3 (only social) or Experiment 4 (only psychology) features can give high accuracy of prediction.
PART B: All features of each domain
The overall accuracy score is lower when no extraction of the Top 10 best features from each domain.
However, Experiment 1 (selecting only demographic features) to predict MMSE and GDS-15 both remain to have the highest accuracy score.
The SVC classifier has the highest average accuracy score of 89.59%, followed by Decision Tree with an average accuracy score of 89.36%, with only a slight difference of 0.23%.
This thus makes the stand of demographic features affecting the prediction of MMSE and GDS-15 the more clearer.
Overall, the prediction of MMSE is much harder compared to GDS-15. The latter prediction can achieve a high average accuracy score with most of the models and combinations except the model Naive Bayes.
Overall, the prediction of MMSE is much harder compared to GDS-15. The latter prediction can achieve a high average accuracy score with most of the models and combinations except the model Naive Bayes.
The best model for both MMSE and GDS-15 prediction is the Decision Tree, with demographic features affecting the most.
The best model for both MMSE and GDS-15 prediction is the Decision Tree, with demographic features affecting the most.
Page updated
Report abuse