Student Academic Prediction Based on Alcohol Consumption Level Using Random Forest Algorithm

Alcohol consumption is a serious social problem, especially among students, due to ease of access and lack of supervision from parents and educators. Alcohol consumption has a negative impact, especially on students, directly or indirectly. Based on a 2015 study, 112 (66%) out of 170 students did not attend classes or failed exams due to alcohol intoxication. It proves that the level of alcohol consumption impacts students’ academic performance. Therefore, it is necessary to predict students’ academic performance to minimize the adverse effects of alcohol consumption. A method to predict academic performance is data mining. This research consists of 5 stages: literature review, data collection, data pre-processing, data processing, and evaluation. The data is obtained from the public dataset "Student Alcohol Consumption", which includes variables such as grades in the first, second and third semesters, demographic aspects, academic aspects, family aspects, social aspects of students, and the alcohol consumption of students on weekdays and weekends. The data pre-processing stage involves several processes, including discretization, set role, and Synthetic Minority Over-sampling Technique (SMOTE) upsampling. The prediction model used to diagnose students’ academic performance was the random forest. The prediction results are evaluated using accuracy, precision, and recall metrics. The results demonstrated that Random Forest produced the best results on the Portuguese language dataset using SMOTE and two-class labels with 95.54% accuracy, 98.08% precision, and 92.90% recall. Based on these results, it can be concluded that the Random Forest algorithm is good for predicting student performance based on their alcohol consumption levels using two-class labels and balanced data. So, this algorithm can help schools mitigate the influence of alcohol on student’s academics.