Predictive Modeling of Anxiety Levels in Bangladeshi University Students: A Voting-Based Approach with LIME and SHAP Explanations

In today's developing world, anxiety is a common mental disorder among university students. In this work, we predict anxiety in university students using a voting classifier. We have applied explainable artificial intelligence (XAI), to gain a better understanding of the machine learning model, using a Google Form, the dataset was gathered from several public, private, and national universities in Bangladesh. We have compared machine learning algorithms using 20 selected features and without feature selection. By using the concept of voting, we have created a new model. In order to create our final model, we selected the best three ML algorithms based on their accuracy. The voting classifier has the highest accuracy of 96%, while F1 and Recall scores are 96%, and Precision is 97%. The LIME and SHAP models are used to explain model predictions instead of a black-box machine learning model. The study determines anxiety levels by particular observations, allowing for transparency and comprehension. The goal is to create prediction models to detect at-risk individuals and root causes of anxiety among university students, consequently reducing detrimental consequences for academic performance and well-being.