A Hybrid Machine Learning Model with Combined Wrapper Feature Selection Techniques to Improve the Yield of Paddy

Abstract

A third of the earth’s surface is taken up by agriculture, which is essential to the food production process. Paddy seeds are used to grow rice, which is a dependable food that is consumed by approximately half of all people worldwide. The alarming rate of population expansion makes it necessary for us to secure food security, and the nation should implement the measures required to increase the production of food grains. Since climatic, agronomic, irrigational, and cultivation techniques all affect paddy’s growth. The goal of the study is to increase the production of rice by using Machine Learning (ML) techniques to forecast the variables that affect paddy growth. More attributes will be employed to build the dataset as ML techniques are used in real-time, which will reduce model performance, raise computing costs, and make the dataset more susceptible to overfitting. This research developed a Hybrid Machine Learning Model with Combined Wrapper Feature Selection Techniques (HMLCWFS) for forecasting paddy production to get over these challenges. The suggested approach selects the most significant features from the Paddy Dataset (PD) using five Feature Selection (FS) approaches: Backward Elimination (BE), Stepwise Forward Selection (SFS), Feature Importance (FI), Exhaustive FS (EFS), and Gradient Boosting (GB) approaches. Using Poincare’s formula, the attributes chosen from each FS approach were concatenated, and the dataset was then recreated. The reconstructed dataset was used to deploy ML approaches like Decision Tree (DT) and Random Forest (RF), and the knowledge gleaned in the form of association rules was utilized to provide advice to paddy growers on how to increase productivity. The suggested model also takes into account the farmers’ preferred paddy farming techniques and makes recommendations regarding which paddy variety they should cultivate. This is accomplished by connecting the input parameters to the real-time PD trained by employing Support Vector Machine (SVM), K-Nearest Neighbor (KNN), and Naive Bayes (NB) methods. The classifier’s results were compared using performance metrics, and the findings demonstrate that the combined FS strategies employed in this research help to identify the elements contributing to the paddy crop’s improvement.