Stacking ensemble learning model for predict anxiety level in university students using balancing methods

Background: Anxiety is known as one of the most common health disorders affecting a large part of the population with a high social and personal impact, which affects about 25% of people worldwide; it is so when it comes to anxiety in students, it is evidenced that in 2018, 63% of high school students in the United States reported having experienced “excessive anxiety” in recent years. Objective: The purpose of this study was to propose a method and 4 combined models based on Stacking with the aim of predicting anxiety levels in college students. In addition, an end-user web interface was developed with the best model proposed in this study. Methods: The data set used consisted of a sample of undergraduate students of systems and computer Engineering from a public university with a total of 284 participants. The data was then cleaned and preprocessed using the Python program. In the data balancing, the data were divided into 5 values obtained and the oversampling method was performed, distributing the data according to the condition. Then the portioning of the balanced data proceeded, using the cross-validation method for data training. For the modeling and evaluation, 5 independent algorithms were used and 4 combined models combined algorithms were proposed. Results: The proposed approach, called Stacking 4A: KNN-Ensemble with data oversampling balancing, was shown to obtain the best results in several evaluation metrics. Specifically, the following values were achieved: Accuracy = 97.83%, sensitivity = 98.44%, f1-score = 97.88%, MCC = 97.08% and specificity = 99.32%, these results exceeded those obtained by the other algorithms. However, the Stacking 2A: SVM-Ensemble technique with data oversampling balance achieved the best value in the precision metric with a result of 97.83%. Conclusions: This article focuses on applying the Ensemble Stacking technique to identify anxiety levels at an early stage among students attending a public university in Peru. Therefore, by using the combined method, an improvement in anxiety prediction was observed, surpassing the performance of the independent algorithms used.