Comparative analysis of data visualization and deep learning models in air quality forecasting

Air pollution stands out as one of the most important environmental problems today. Factors such as increasing population, rapid urbanization, and industrialization lead to increasing air pollution levels. In our country, pollutants in the air are measured regularly in different regions and the data obtained from these measurements constitute an ever-growing data set. However, the increasing data volume also brings some difficulties. The variety and size of the data collected from different sources make the analysis and modeling processes more complex. In addition, the movement of pollutants in the atmosphere and the effects of environmental variables make it even more difficult to estimate air quality accurately. In this study, using the data provided by the Continuous Monitoring Center of the Ministry of Environment, Urbanization and Climate Change of Turkey, the data of different pollutants in the air were estimated comparatively with three different deep learning methods, namely Convolutional Neural Network (CNN), Recurrent Neural Network (RNN) and Long Short-Term Memory (LSTM), the prediction of the next three years with the method that gave the best results in the prediction and then the visualization of these obtained values were provided. Violin Plot, Box Plot, and Point Scatter graphics were used as data visualization. It is thought that data estimation will be useful in reducing future uncertainties. Data visualization makes it easier for non-experts to estimate and understand air quality information from the displayed concentration profiles. Among the models used in this study, accuracy rates of 0.88 for PM10 with CNN, 0.93 for SO2, 0.94 for PM10 with LSTM, and 0.95 for SO2 were achieved. This thesis study includes comparative analyses of deep learning models in air quality estimation and data visualization processes using air pollution data from Başakşehir district of Istanbul. The originality of the study is that both visualization and deep learning methods are considered together.