Improving student experience with a flexible machine learning system for tracking air quality

Abstract

Air pollution is significantly a significant silent warning for human life in this digitized era. This pollution is a horrible threat for public health. Students spend most of the day at their educational premises, so the air quality of any educational setting has a wide impact on students’ health, which is a great concern. This research focuses on building a hybrid model combining machine learning and deep learning techniques to predict Air Quality Index (AQI) levels more accurately than existing models. The dataset consists of 1632 data points with key pollutants Carbon Dioxide (CO2), Carbon Monoxide (CO), Total Volatile Organic Compounds (TVOC), Formaldehyde (HCHO), Particulate matter 0.3 (PM0.3), Ultrafine particulate matter(PM1.0), Fine particulate matter (PM2.5), and Particulate Matter (PM10.0) collected from 18 different locations on campus over 7 months (January 2025 to July 2025).Multiple machine learning (ML) models, e.g., Random Forest (RF), eXtreme Gradient Boosting (XGBoost), Categorical Boosting (Cat- Boost), Light Gradient Boosting Machine (LightGBM), and Support Vector Machine (SVM), as well as deep learning (DL) models, including Gated Recurrent Unit (GRU), Long Short-Term Memory (LSTM), and a proposed hybrid model, were trained and evaluated using standard metrics such as accuracy, precision, recall, F1- score, confusion matrix, and ROC-AUC curve after appropriate data preprocessing. Additionally, as a part of explainability integration, SHapley Additive explanations (SHAP) were observed for the proposed model in this study. The forced Shap plot for different samples in each of the five classes of AQI classification both correctly and incorrectly predicted the class to detect the feature contribution behind the model’s prediction. Experimental results show that the proposed hybrid model accuracy is highest among all used machine learning and deep learning models. CatBoost, LightGBM, and Random Forest are the top three models in machine learning, and SVM is the highest in deep learning models. The proposed model air quality classification system effectively combines machine learning with explainable AI methods to improve experience and build trust in environmental monitoring. It shows promising potential for real-time use in educational settings to increase awareness and enable informed health decisions. However, this study has few limitations, e.g.an imbalanceded dataset, lack of data in each class, and model’s limitations, e.g., overfitting risk, lack of tuning, and so on; future efforts will concentrate on real-world validation and refining the model.