Feature Selection Technique to Improve the Instances Classification Framework Performance for Quran Ontology

The Al-Quran is the sacred book of Muslims, and it provides God's word in the form of orders, instructions, and guidelines for people to follow to have happy lives both here and in the afterlife. Several earlier research has used ontologies to store the knowledge found in the Quran. The previous study focused on extracting the relationship between classes and instances or the "is-a relation" by classifying instances based on the referenced class. Based on the performance testing of the instances classification framework, the test results show that Support Vector Machine (SVM) with Term Frequency-Inverse Document Frequency (TF-IDF) and stemming operation had dropped the accuracy value to 65.41% when the test data size was increased to 30%. Likewise, with BPNN with TF-IDF and stemming operations. In the Indonesian Quran translation dataset with a test data size of 30%, the accuracy value drops to 57.86%. Instances classification based on the thematic topics of the Qur'an aims to connect verses (instances) to topics (classes) to get an overall picture of the topic and provide a better understanding to users. This study aims to apply the feature selection technique to the instances classification framework for the Al-Quran ontology and to analyze the impact of applying the feature selection technique to the framework with a small dataset and training data. The instances classification framework in this study consists of several stages: text-preprocessing, feature extraction, feature selection, and instances classification. We applied Chiq-Square as a technique to perform feature selection. SVM and BPNN as a classifier. Based on the experiment results, it can be concluded that the feature selection implementation using Chi-Square increases the value of precision, f-measure, and accuracy on the test data size from 40% to 60% in all datasets. The feature selection using Chi-Square and SVM classifier provides the highest precision value with a test data size of 60% on the Tafsir Quran dataset from the Ministry of Religious Affairs Indonesia: 64.36%. Furthermore, the feature selection implementation and BPNN classifier also increase the highest accuracy value with a test data size of 60% in the Quranic Tafsir dataset from the Ministry of Religion of the Republic of Indonesia: 63.09%.