Residence state and country prediction of student towards ict for the real-time

An experimental study was conducted to predict the residence state and country of students based on their response provided in the two different ICT survey held during the academic year 2015-2016 and during the academic year 2017-2018. The first dataset was consisted of 560 instances and 59 features and second dataset was comprised of 331 instances and 46 features. We considered the state in the first dataset and country in the second dataset as the response variable and rest of all are assumed as predictors after self-reduction few features. The datasets are trained and tested with the splitting and k-fold Cross Validation (CV) using three popular supervised machine learning classifiers named Artificial Neural Network (ANN), Sequential Minimal Optimization (SMO) and Random Forest (RF) in the Weka 3.8.1 workbench. In the state prediction, the RF classifier outperformed with highest prediction accuracy of 83.39% the ANN and SMO at 6-Fold of the CV method. The maximum accurate prediction count for the Punjab student is found 239 out of 282 and for the Haryana student is found 228 out of 278 with k=6. In the country prediction, the best fitting model will be presented with highest prediction accuracy. The comparison findings are described about state versus country prediction of student with important measures like accuracy, error, F-score, Confusion matrix, True Positive Rate (TPR), False positive rate (FPR) and ROC curves. Further, these state and country predictive models may support the real-time prediction module of student's demography prediction towards the technological awareness.