現今社會隨著工業化和都市化加速的同時，污染物經由汽車排放、工業生產和能源消耗等過程釋放至大氣中，其造成空氣污染已成為一個嚴重的全球性問題，而參與AQI評價的污染物不僅造成空氣方面的污染，對人體健康影響更如同是無形的毒藥。 透過機器學習模型之運用，建立一個準確預測空氣污染物濃度的模型與訓練，以期能夠對未來的空氣品質進行預測分析。模型訓練上搭配線性回歸分析來探討空氣污染物的相關性，並採用梯度下降法作為資料收斂的最優化演算法，以最小化目標函數及達成模型之預測。 以不同數據資訊之組合進行預測分析訓練，另考慮了各地區對於氣候及污染物間的相關性與可預測性之可能差異，故亦分別依照地區選擇不同的測站作分析驗證比較。本論文以空氣中二氧化硫濃度作為預測值，來探究氣候條件與空氣污染物之間的相關性。選擇了適當的特徵值進行預測分析，探究了模型系統預測值與特徵值之間的相關性與可預測性確認，最後建立一個有效之預測模型，達到最佳之預測成效。 ;Nowadays, due to the acceleration of industrialization and urbanization, pollutants are released into the atmosphere through several activities such as vehicle emission, industrial production, live wastes disposition and energy consumption. Air pollution has become a serious global problem. The pollutants involved in the AQI evaluation cause air pollution, also act like the invisible poisons to human health. By model training with machine learning, a model for accurately predicting the concentration of air pollutants was established to analyze the air quality and predict the future status. The model training was combined with the linear regression analysis to explore the correlation and prediction model of air pollutants. The gradient descent method was used as the optimization algorithm in for data convergence in order to minimize the objective function in this study. The model training combines the linear regression analysis to explore the correlation between the air pollutants and uses the gradient descent as the optimization algorithm for data convergence to minimize the objective function and achieve the prediction of the model. Prediction analysis and training was conducted using different combinations of data and information. In addition, considering the differences of correlation and predictability may exist between the climate and pollutants in various regions, in the study, select ed different measuring stations according regions as the basis for comparison and verification. This study selected the sulfur dioxide in the air as the prediction value to explore the correlation between the climate conditions and the air pollutants. Selected the appropriate feature values to conduct the prediction analysis, explored the correlation and the predictability between the prediction values of the model system and the feature values. Finally, established an effective prediction model and achieved the best prediction results for the air quality.
