Prediction of residence country of student towards information, communication and mobile technology for real-time: preliminary results

Abstract

An experimental study is conducted to solve a binary classification problem, and to predict the country of residence (India and Hungary) for the students towards Information Communication and Mobile Technology with various parameter. Four supervised machine learning algorithms such as Random Forest (RF), K-nearest Neighbours (KNN), Multilayer Perceptron (ANN) and Support Vector Machine (SMO) are modeled in Weka tool on primary dataset (331 instances & 46 attributes) with the four parameters Attitude (A), Development and Availability (DA), Usability (U) and Educational Benefit (EB) with Feature Selection. Initially, these algorithms are applied on individual datasets and later, these are applied on Aggregate (AG) dataset. The datasets are trained and tested using three different type of testing methods such as Leave One Out, Hold Out and K-Fold Cross Validation (CV). Therefore, a statistical T-test at 0.05 significance level is also used to compare the prediction accuracy of classifiers on individual datasets and accuracy are measured. Afterward, Feature Selection methods such as ClassifierSubsetEval (CSE), GainRatioAttribu-teEval (GRAE) and InfoGainAttributeEval (IGAE) are also applied to enhance the prediction accuracy of AG-37 dataset. Initially, with T-test, the RF classifier with the Leave One Out testing method outperformed others. It is also explored that the CSE filter significantly improved the prediction accuracy of KNN, ANN, and SMO using filtered out 8, 14 and 15 features respectively. Further, 10 ranked features by GRAE filter significantly enhanced the accuracy of CSE filter for the KNN classifier and also for the RF classifier on AG-37. Latter, IGAE filter also boosted the prediction accuracy of the KNN, SMO and RF with 11 significant features. Also, the highest prediction accuracy 91% is gained by SMO with CSE-15 features, 90.3% accuracy is gained by RF with GRAE-10 features and 92.8% accuracy is gained by RF with IGAE-11. Finally, to predict the country of residence of the students towards Information Communication and Mobile Technology, the RF classifier outperformed others using Feature Selection in optimum time (0.14 seconds).