Enhancing Cotton Crop Yield Prediction Through Principal Component Analysis and Regression Modelling

The prediction of cotton crop output is an essential undertaking in the field of agriculture, as it provides significant knowledge regarding forthcoming harvests and empowers farmers to make well-informed choices. This research integrates two robust methodologies, namely PCA and OLS regression, in order to effectively forecast cotton crop yields. The dataset consists of yearly observations pertaining to the production of cotton crops from 1964 to 2015. These observations are presented in the form of time series. In addition, the variables from (Lag1 to Lag5) representing the historical yield values of India are included in the dataset. Principal component analysis (PCA) is used to determine the covariance matrix, eigenvalues, eigenvectors, and projected data values that are studied in this research. This research provides the fundamental framework of the dataset and aids in identifying the important components to be utilized in forecasting. Principal components are recognized and has been used in an ordinary least squares regression analysis taking the cotton crop yield as the dependent variable. The analysis shows a strong predictive correlation between the identified primary factors and the production of cotton crops. Root mean square error (RMSE) and R2 are the metrics used to evaluate the effectiveness of regression models. The results show the effectiveness of combining principal component analysis (PCA) with ordinary least squares (OLS) regression to calculate the cotton crop yield accurately. The above method shows promising results in improving the forecasting of cotton crop yield and can be further extended; thus helps in the advancement of sustainable agriculture and guarantees food security. Based on this study, the OLS regression model helps in predicting the yield of cotton crop when it is combined with the PCA-1. The model shows the performance on both the training and testing datasets producing root mean square error (RMSE) values that are comparatively low.