Analysis of a Novel Integrated Machine Learning Model for Yield and Weather Prediction: Punjab Versus Maharashtra

Abstract

Listen

Translate article

Correct prediction is a deciding factor in modern agricultural techniques to confirm food safety and sustainability faced in crop production. As radical climatic conditions affect plants growth greatly, appropriate valuation of rainfall and yield prediction can deliver a sufficient information in advance which can be utilized to maintain the crop production quality. For accurate prediction, machine learning (ML) plays a prominent role. The collaboration of agriculture system with machine learning will lead to intelligent agriculture system that helps the farmer community in their decision-making of farm management and agribusiness activities such crop management including applications on yield prediction, disease detection, weed detection, crop quality, and growth prediction. This paper implemented and analyzed a prediction model using high potential ML algorithms, viz., random forest, linear regression, Lasso regression, and support vector machine for crop yield prediction whereas KNN, decision tree, logistic regression, Gaussian Naive Bayes, SVM, and linear discriminant analysis to find the best line method for rainfall prediction. To enhance the effectiveness of pre-processing, “SSIS-an ETL” tool is utilized. The experiments laid out for the state of Maharashtra and Punjab, and dataset collected from “ICAR-Indian Agriculture Research Institute” and www.imd.gov.in . R2-score, mean squared error, precision, recall, and F-score measure were used to ascertain the accuracy. As a result, random forest found significantly the finest algorithm for crop yield prediction by recording higher accuracy of 96.33% with maximum R2-score of 0.96 and least mean squared error (0.317), while lasso regression was found the worst with an accuracy of 32.9%. In case of rainfall prediction, Gaussian Naive Bayes secured top rank by recording considerably the highest accuracy of 91.89% as well as the maximum precision, recall, and F1-score of 0.93, 0.91, and 0.91, respectively.