A novel analysis and prediction of students' behaviour using semantic similarity-based improved J48 IL algorithm in personalized library ontology

Abstract

The Semantic web deliberates machines to develop conceptual information themselves by understanding its implications. Semantic web creates overt resources through ontology. Ontology widely offers the representation of conceptual knowledge. It has the ability to signify the domain knowledge in a distinct and unambiguous manner. We propose Ontology-based structure for personalized library usage environment which discover user's behaviour by their learning feedback. The personalized library Ontology and user's feedback is developed using Protégé editor 4.3. To identify user behaviour we determine 225 students' library learning feedback which enhance the reference information for the future users. For data classification, we propose semantic similarity-based Improved J48 induction learning algorithm which facilitates the system to identify patterns and regularities for the extracted ontology data. Impurity from the classified data is evaluated using entropy and information gain. The new decisionmaking for ontology subpopulation is made by estimating the highest GainRatio. Finally, the system performance is evaluated with k-fold cross-validation, accuracy, F1-score, and execution time. The proposed algorithm reduces dissimilar data and handles missing data therefore, the system can improve its efficiency and performance with enhanced accuracy up to 10% - 12% compare to existing algorithms as Random Forest, Naive Bayes, Multilayer Perceptron and J48 algorithm. It can also reduce its execution time more than 20 (ms) for the different size of datasets compare to existing C4.5 algorithm. Theoretical analysis and comparison of existing algorithms with its experimental results and system evaluation show the effectiveness and performance of the proposed system.