Predicting the impact of internet usage on students' academic performance using machine learning techniques in Bangladesh perspective

Education systems have significantly changed with the emergence of the internet. It has a significant impact on how students learn things. Nevertheless, its impact can also be contradicting. Internet addiction can slowly poison the minds of our youths and stand in the way of pursuing their goals. Although Bangladesh has internet connectivity across the country, its potential could be more utilized, particularly in the educational sector. Therefore, proper analysis of the effects of the internet on students, as well as determining the prominent factors relevant to the internet, is a necessary task. In addition, predicting students' academic performance can help determine the changes that must be incorporated to improve the educational system. Hence, this research analyzes the effects of internet usage on students' academic progress and then predicts the students' performance using distinct machine learning (ML) algorithms. The data were collected through an offline survey from Noakhali, Bangladesh. The collected data is preprocessed to select the most relevant features. The preprocessed data were fed into ML algorithms to investigate their behaviors. We have employed logistic regression, decision tree, random forest, and Naive Bayes algorithms to see their classification performance on our dataset. To minimize the overfitting issue, k-fold cross-validation and hyperparameter optimization have been applied. The results were presented in two parts-exploratory data analysis and classification. Exploratory data analysis shows that the main purpose of internet usage is education and entertainment for school students, social media and entertainment for college students, and education and social media for university students. School and university students browse the internet mainly for academic purposes, whereas college students browse mainly for non-academic purposes. Students prefer to browse the internet at night. For all schools, colleges, and universities, students with better results generally visited websites like Google and YouTube. Students with moderate or bad results generally spent time on social media platforms (mainly Facebook and WhatsApp). Then, the results of the numerical analysis performed with classification algorithms are presented. Results indicate that random forest gives the maximum score in our dataset in all sectors, like accuracy, precision, recall, and f1 score. It gives a maximum of 85% accuracy on the test set. Logistic regression gives the second-best score of 69%. The practical applications and policy recommendations for Bangladesh's education sector are also discussed. The output of this work can contribute to building a policy on internet usage. In this way, it is possible to make the students more concentrative on their education and learning.