Cascading ensemble machine learning algorithms for maize yield level prediction

Seireg, Hayam R; Omar, Yasser M.; El-Sayed, Fathi; El-Fishawy, Adel Shaker; Elmahalawy, Ahmed

doi:10.21608/mjeer.2023.159995.1066

Cascading ensemble machine learning algorithms for maize yield level prediction

Document Type : Original Article

Authors

¹ Department of Computer Science and Engineering, Faculty of Electronic Engineering, Menoufia University, Menouf , Egypt

² College of Computing and Information Technology Arab Academy for Science, Technology and Maritime Transport (AASTMT) Cairo, EGYPT

³ Department of Electronics and Electrical Communications, Faculty of Electronic Engineering, Menoufia University, 32952, Menouf, Egypt.

⁴ Electronics and Electrical Communications Engineering Dept., Faculty of Electronics Engineering, Menouf, Menoufia University, EGYPT

⁵ Department of Computer Science and Engineering, Faculty of Electronic Engineering, Menoufia University, Menouf, EGYPT

10.21608/mjeer.2023.159995.1066

Abstract

Climate change is destroying many crops around the world. This paper aims to anticipate maize yield levels based on climatic conditions, which would aid in making proper decisions regarding the connected sectors for business planning and yield level prediction. This paper presents two novel models that combine five machine learning algorithms with different techniques. Selecting six months of the climate features for the four regions in China. The first proposed model (FPM) consists of K Nearest Neighbors, Multinomial Naïve Bayes, Bernoulli Naïve Bayes, Decision Tree and Quadratic Discriminant Analysis (KMBDQ) that come together in a cascading topology (CT) to feed each other by taking the new prediction and removing the old previous prediction from the input features at each stage. The second proposed model (SPM) uses the same mentioned algorithms with different approaches. The prediction of each machine learning (ML) is used as a feeder to each other in the form of CT without removing any prediction. The performance evaluation of the proposed models was demonstrated and compared with many classifiers with the same dataset using accuracy, sensitivity, precision and F1 score. The results revealed that the SPM had the highest prediction accuracy of 79.6% with an increase of 29.6% compared to the first classifier in the model. It also had an improvement of 11.1% than the FPM and an increase of 10.2% compared to the best one among the many techniques used. Moreover, computation time comparisons are spotted.

Keywords