The Study of Features Affecting the Digital Literacy Test and Comparative Efficiency of Data Classification Using Data Mining Techniques

The purpose of this research is to study the features that affect the results of the test for Digital Literacy by using the Feature Selection technique to compare the efficiency of Data Classification. The research process follows the steps of CRISP-DM, studying and collecting 12,374 records of 67 features of the Office of the National Digital Economy and Society Commission's survey and assessment of the state of digital literacy. Carry out Data Cleaning and Data Transformation into an appropriate format. SMOTE (Synthetic Minority Oversampling Technique) was used to improve the data balance. Features were selected using weighting calculation techniques: 1) Chi-Square Statistic 2) Gini Index 3) Gain Ratio 4) Information Gain and 5) Correlation-based. Using the results of the first 10 calculated highest weights of each technique to calculate the frequency and Features with frequencies higher than 2 were selected to design and create forecasting models of 5 techniques: Decision Tree, Gradient Boosted Trees, Random Forest, Naïve Bayes, and Deep Learning. Evaluation of the forecasting model using the K-Fold Cross Validation Test method, dividing the data into 10 folds, 20 folds, and 30 folds, measuring precision, recall, specificity, accuracy, F1-measure, and G-means. The results of the Study Feature Affecting the Digital Literacy Test revealed that there were 6 features with the highest frequency, frequency 5, as follows: the use of advertising media/product labels, using digital media for social media, using digital media to access websites, the problem of expensive internet service fees, problems accessing internet service areas and problems with spam/advertising messages. The comparative results of data classification efficiency test results of digital literacy found that the random forest technique was the most effective in data classification. When dividing the dataset into 30 parts, the accuracy was 76.29%, the overall efficiency was 76.00%, and the geometric mean was 76.28%.