International Research Journal of Engineering and Technology (IRJET)
e-ISSN: 2395-0056
Volume: 13 Issue: 06 | Jun 2026
p-ISSN: 2395-0072
www.irjet.net
A HYBRID APPROACH FOR PREDICTION OF AIR QUALITY INDEX USING MACHINE LEARNING Sarvesh Garg1, Dr. Kiran Jyoti2, Dr. Hardeep Singh Kang3 1Student & 1765, Phase-1, Urban Estate, Dugri. Ludhiana 2 Professor, Dept. of Computer science Engineering, GNDEC, Punjab, India
3 Associate Professor, Dept. of Computer Science Engineering, GNDEC, Punjab, India ---------------------------------------------------------------------***---------------------------------------------------------------------
Abstract - Air pollution has become one of the most critical
data and advances in computational power, data-driven methods have gained significant attention. Machine learning and deep learning models, such as artificial neural networks, convolutional neural networks, and recurrent neural networks, have demonstrated superior predictive capabilities compared to classical models. Nevertheless, single deep learning models may still face limitations related to feature extraction, temporal dependency modeling, spatial correlation handling, and generalization. These challenges have motivated the exploration of hybrid deep learning methods for air quality prediction.
environmental and public health concerns worldwide due to its severe impact on human health, ecosystems, and climate conditions. Rapid urbanization, industrialization, and increasing vehicular emissions have significantly contributed to the deterioration of air quality, making accurate forecasting of air pollution levels essential for effective environmental management and policy planning. Predicting the Air Quality Index (AQI) with high precision is challenging because air pollution data exhibit both nonlinear patterns and temporal dependencies influenced by meteorological conditions, pollutant interactions, and seasonal variations. Traditional statistical models often fail to capture nonlinear relationships effectively, whereas standalone machine learning models may not fully address time-series dependencies and residual errors. To overcome these limitations, this research proposes a novel hybrid machine learning framework for air quality prediction by integrating Random Forest (RF), Autoregressive Integrated Moving Average (ARIMA), and Gated Recurrent Unit (GRU) models. The proposed methodology combines the strengths of these three models to improve prediction accuracy and robustness. Initially, the Random Forest algorithm is employed as the primary predictive model due to its strong capability in handling nonlinear relationships, feature interactions, and highdimensional environmental datasets.
2. LITERATURE REVIEW S. et. al, 2025 said that the research employed an LSTM model to predict the Air Quality Index (AQI) from a threeyear dataset of eight pollutants. The data were smoothed and normalized prior to training with look-back days as one of the important parameters. Experimental results had best performance with MAE 0.032 for training and 0.043 for test at 10 look-back days [1]. Khokhlov et. al, 2025 proposed that this research used ML and DL models to forecast AQI in Tashkent, Uzbekistan, one of the dirtiest cities in the world. Using KNN, RF, DT, SVM, and ANN using optimizers such as Adam, Ada Grad, RMSprop, and SGD with momentum, the problem was set as classification. Model performance was measured using Accuracy, Precision, Recall, and F1-Score, identifying strengths, weaknesses, and appropriate methods to predict AQI [2].
Key Words: Deep Learning, Gated Recurrent Unit, Elastic Net, Machine Learning, Air Quality.
1. INTRODUCTION
Au et. al, 2025 said that the research used Automated Machine Learning (Auto ML) on time- series IoT sensor data in air quality monitoring. Experiments revealed that Auto ML models had similar or better accuracy compared to customary and neural network algorithms such as Random Forests, Extra Trees, Elastic Net, and LSTM. The findings exhibited the usability and cross-industry applicability of ML-based air quality analysis without the need for specialized technical knowledge [3].
Air quality has become a critical environmental and public health concern due to rapid urbanization, industrialization, and increased vehicular emissions. Poor air quality, characterized by high concentrations of pollutants such as PM₂.₅, PM₁₀, NO₂, SO₂, CO, and O₃, has been directly linked to respiratory diseases, cardiovascular disorders, and premature mortality. To effectively mitigate these impacts, timely and accurate air quality prediction is essential. Traditional statistical and deterministic models, including linear regression and chemical transport models, have been widely used for air quality forecasting. However, these approaches often struggle to capture the highly nonlinear, dynamic, and complex interactions among meteorological variables, emission sources, and atmospheric processes. With the availability of large-scale air quality monitoring
© 2026, IRJET
|
Impact Factor value: 8.315
Kocak et, al, 2025 said that this research compared five machine learning models to predict hourly PM2.5 and PM10 concentration based on actual pollutant and meteorological data. Varied performance was observed, with the best performance in PM2.5 prediction by SVR (R²=0.83), followed
|
ISO 9001:2008 Certified Journal
|
Page 898