Rainfall Prediction and Suitable Crop Suggestion Using Machine Learning Prediction Algorithms

Abstract

Farmers in India suffer gruesome fate at the mercy of rain gods since primary source of agriculture in India is rainfall. Agriculture is a major source of living but contributes only about 18% of total gross domestic product, its reason being lack of adequate crop planning by farmers. Although India has surplus fertile land, inefficient agricultural practices due to deficiency of rainfall and crop prediction techniques, in turn, leads to uncountable farmer suicides. Currently, the invasion of Machine Learning (ML) has abetted in finding promising solutions to address the problems of predicting rainfall, soil assessment, crop management, yield prediction, crop quality and disease detection and classification. Despite the technology, hitherto, there is no platform nor system in place to inform farmers of the rainfall predicted and advise what crops to grow. In this paper, rainfall prediction using diverse ML and statistical algorithms is encapsulated, accordingly best suitable crops to grow are recommended keeping soil as a parameter. The raw real-time rainfall data acquired was pertained to three regions of Karnataka North, South and Coastal. Data was cleaned and structured and its features extracted. Statistical tests—ADF, KPSS, ACF, PACF executed on the feature extracted data revealed its trend and seasonality insightful for modelling. Using effective ML and statistical algorithms such as ARIMA, ANN, random forest, TBATS, Holt-Winters, simple, double, triple exponential smoothening et al., rainfall for the next six years was predicted. All three regions were distinctly modelled. Time series forecasting using ARIMA proved to be the best performer. All models performances are validated using standard error measures to have authenticity of accuracy. The generalized accuracy of ARIMA model averaging on all three regions is 92.91%, ANN 88.26%, TBATS 61%, simple exponential smoothening had 71.1%, double exponential smoothening is 68.63%, triple exponential smoothening accuracy is 57.42%, and random forest gave 42%.