A Discussion on Machine Learning Approach of Rainfall Prediction

A comprehensive correlation analysis of extreme rainfall (>204.5 mm/day) and very heavy rainfall (> 115.6 mm/day) has been done over Indian subcontinent considering 40 years (1981–2021) of data of meteorological parameters having a causal relation with rainfall. Seasonal analysis has been done after locating the place having maximum rainfall over the forty years for both the two categories of extreme rainfall and very heavy rainfall. For each of the categories best correlated parameters are taken for prediction of rainfall through linear regression, a machine learning based approach. The performance analysis of the prediction has been done. In the next step another machine learning approach, multivariate linear regression analysis has been done connecting rainfall with the associated parameters. Performance analysis of multivariate regression also has been done for rainfall prediction. Finally, comparison of performances for single variable analysis and multivariate analysis has been done to come to a conclusion about the prediction of rainfall (Precip) over Indian landmass and it has been found that best correlation performance has been obtained at the month March-April-May at Acharkund, Madhya Pradesh (22°30′00.0″N 79°00′00.0″E) taking the wind speed $(\mathrm{U}_{2})$, relative humidity(RH), specific humidity(SH), surface temperature(T), dew point temperature $(\mathrm{T}_{\text{dew}})$ and surface pressure(P) as input variable for prediction.