Comparative Analysis of Data Visualization and Deep Learning Models in Air Quality Forecasting

This study utilizes air pollution data from the Continuous Monitoring Center of the Ministry of Environment, Urbanization, and Climate Change in Turkey to predict various pollutants using three advanced deep learning approaches: LSTM (Long Short-Term Memory), CNN (Convolutional Neural Network), and RNN (Recurrent Neural Network). Missing data in the dataset were imputed using the K-Nearest Neighbor (K-NN) algorithm to ensure data completeness. Furthermore, a data fusion technique was applied to integrate multiple pollutant enhancing the richness and reliability of the input features for modeling. The increasing air pollution issue, driven by factors such as population growth, urbanization, and industrial development, is a major environmental concern. The study evaluates these models to estimate pollutant concentrations and selects the most accurate, RNN, for forecasting air pollution over the next three years. Each prediction was assessed using performance metrics such as MAE, RMSE, and R² to ensure robust model evaluation. Visualization of the data and forecast results was achieved through methods like Box Plots, Violin Plots, and Point Scatter Graphs, making air quality information more accessible to general audiences. In terms of model performance, CNN achieved an R² of 0.88 for PM10 and 0.93 for SO2, while LSTM demonstrated an R² of 0.94 for PM10 and 0.95 for SO2. However, RNN emerged as the most accurate model, achieving an R² of 0.97 for both PM10 and SO2 forecasts. This model allows for forecasts of pollutant levels over a three-year period. The findings indicate that predictive modeling, combined with data fusion and visualization techniques, could significantly contribute to mitigating future uncertainties and enhance the comprehension of air quality patterns for non-expert audiences.