Machine Learners’ Performance Analysis for Meta-Learning Base Classifier Selection in Plant Disease Diagnosis

Abstract

Efficient hyper-parameter tuning is crucial for high-performing machine learning in plant disease diagnosis. This study comparatively analyzes K-nearest neighbor, random forest, neural networks, and AdaBoost models under a developed meta-learning Machine Learners’ Performance Analysis framework to optimize their hyper-parameters for cassava plant disease datasets. We evaluate their adaptability, efficiency, and robustness using metrics like accuracy and computational cost. Applying meta-learning techniques, including gradient-based optimization and neural architecture search, we identify significant performance variations. Findings underscore the importance of selecting the optimal model-meta-learning combination for effective and automated plant disease diagnosis in precision agriculture.