Performance Comparison of Data Mining Techniques for Building Classification Models of the Parent Toward Children who use Smart Phone

Abstract

This research aimed to compare the efficacy of a data mining technique in modeling parental opinion on the use of their children's smartphones. This research investigated opinion mining, comparing the model performance of 6 techniques: the RIPPER, C4.5 decision tree,  Naïve Bayes,  Support Vector Machine,  K-Nearest Neighbor,  and Random Forest technique to analyze the opinions of parents on the use of smartphones of their children. Therefore, the data was collected from only the Thai texts of parents' opinions on the social platforms: Pantip and Facebook,  a total of 1, 925 messages. In this process, only adverbs were used to classify the state of the word,  conveying both positive and negative sentiments well. The researcher applied the 10-fold cross-validation performance measurement technique to cluster the learning data and the testing data and measure the model performance using Precision, Recall,  and F-measure. After testing and measuring the performance of the model, it could be concluded that the Random Forest technique was the best technique to analyze the opinions for this data set. Conclusively,  this Random Forest technique provided a model performance score of 83.85% with Precision of 89.62%,  Recall of 78.38%,  and F-measure of 83.55%.