A Comparative Study of Statistical and Machine Learning Modelling Techniques in Air Pollution Data

Abstract

Different approaches are being adopted, in practice, for determining models for given time series. The approaches can be categorized broadly into three viz., statistical, machine learning and deep learning. Since they differ with respect to their theoretical base, their outcomes also differ. Decision making based on the values predicted from the time series models seek accuracy of the forecast values.  This paper studies the effectiveness of the three approaches by comparing the performance of the autoregressive moving average method developed applying statistical principles, Facebook Prophet method developed from Machine Learning approach and long short-term memory method developed from deep learning.  The study is carried for real data of time series of air quality indices.