Comparison of multiple linear regression and multiple nonlinear regression models for predicting rice production.

Rice is the staple food for Asia and a part of the Pacific, including Malaysia. However, the self-sufficiency ratio (SSR) for rice production in Malaysia has dropped from 69% in 2019 to 65% in 2021. For the continuous prosperity of Malaysia in food security, therefore this paper aims to investigate the statistically significant factors that affected the reduction of rice production based on the Cross Industry Standard Process for Data Mining (CRISP-DM) data science methodology. To pursue the principal objective of this paper, the annual rice production dataset period 1980-2019 corresponding to the atmospheric, climatic, and socio-economic factors under big data has been employed. Meanwhile, the regression-based predictive models employed in this paper are including multiple linear regression (MLR) and multiple nonlinear regression (MNLR) supervised machine learning models. The empirical analysis revealed that the MLR is superior rather than the MNLR in predicting rice production. Both supervised machine learning models consistently showed that the planted area and annual population are the statistically significant factors for the supervised machine learning models. In summary, this study is competent to support the advancing sustainability theme via the principal focus of poverty income revision. Specifically, the superior parsimonious predictive model resulting in this study is competent to beneficial for smallholder farmers in adopting plantation strategies for the future. Furthermore, this predictive model also could be beneficial to policymakers by providing early alarm insight about the national impact of recent atmospheric, climatic, and socio-economic trends on rice production. [ABSTRACT FROM AUTHOR]