Browsing by Author "Dauletkhan N."

Now showing 1 - 1 of 1
  • Thumbnail Image
    ItemOpen Access
    A comparative study of air quality analysis in Almaty
    (SDU University, 2025) Dauletkhan N.
    Air pollution remains a pressing public health and environmental challenge in Almaty, Kazakhstan, where concentrations of fine particulate matter (PM2.5) frequently exceed World Health Organization limits. This study presents a comprehensive comparative analysis of statistical, machine learning (ML), deep learning (DL), and hybrid models for short-term PM2.5 forecasting using real-world meteorological and air quality data collected between 2020 and 2024. The methodology involved rigorous data preprocessing, including imputation techniques such as mean substitution, time-based mean, and Multiple Imputation by Chained Equations (MICE), followed by correlation analysis and normalization. Multiple models were implemented and evaluated: statistical models like Multiple Linear Regression (MLR), SARIMA, and Prophet; ML algorithms including Random Forest, Support Vector Regression (SVR), and XGBoost; DL architectures such as Long Short-Term Memory (LSTM) and Convolutional Neural Networks (CNN); and hybrid combinations like CNN-ELM and CNN-LSTM. Model performance was assessed using MAE, RMSE, and R² across three imputation scenarios. Results indicated that LSTM consistently achieved the highest accuracy, particularly under the MICE imputation scenario, while Random Forest and XGBoost showed strong performance among ML models. Hybrid models like CNN-LSTM demonstrated promising results in capturing both spatial and temporal patterns. This research contributes to the development of robust, interpretable, and localized forecasting systems, offering valuable insights for environmental monitoring and public health planning in data-constrained urban regions