Reducing the Error Mapping of the Students’ Performance Using Feature Selection
2021
In an educational environment, classifying the cognitive aspect of students is critical. It is because an accurate classification is needed by a lecturer to take the right decision for enhancing a better educational environment. To the best of our knowledge, there is no previous research that focuses on this classification process. In this paper, we propose discretization and feature selection methods before the classification. For this purpose, we adopt the equal frequency for the discretization whose result is evaluated by using logistic regression with two regularizations: lasso and ridge. The experimental result shows that four-intervals on the ridge achieve the highest accuracy. It is to be the base to determine the level of the student’s performance: excellent, good, fair, and poor. Next, we remove unnecessary features, by using the Gain Ratio and Gini Index. Also, we build classifiers to evaluate our proposed methods by using k-Nearest Neighbors (k-NN), Neural Network (NN), and CN2 Rule Induction. The experimental result indicates that both discretization and feature selection can enhance the performance of the classification process. Concerning the accuracy level, there is an increase of about 35%, 2.14%, and 3.8% on average of k-NN, NN, and CN2 Rule Induction respectively, from those with original features.
- Correction
- Source
- Cite
- Save
- Machine Reading By IdeaReader
24
References
0
Citations
NaN
KQI