Predicting PM2.5 Value in Future

• Kaushal Thaker

  Author

PM2.5 pollution poses a dangerous threat to human health and the environment; therefore, accurate forecasting methods are essential. This paper applies machine learning techniques to predict future PM2.5 levels using time series data. Two prediction tasks are considered: (i) regression to predict short-term PM2.5 values, and (ii) classification to determine the air quality level for the next day. Historical air quality data from several Chinese cities (primarily Beijing) are used to train and test neural networks and regression models. The neural network achieves a correlation coefficient exceeding 0.95 for the regression task, indicating that recent past data serve as strong predictors of near-future PM2.5 values. The classification task proves more challenging, achieving approximately 50\% accuracy due to class imbalance and limited data. Ablation studies show that local temporal patterns suffice for short-term prediction, whereas more complex tasks demand richer features and models. The paper further extends the approach into an enterprise-ready predictive analytics framework, including system architecture, data pipeline design, model comparison, validation strategies, and real-world impact assessment.

• PDF

Copyright (c) 2026 International Journal of Intelligent Systems and Data Science

This work is licensed under a Creative Commons Attribution 4.0 International License.