Cassava leaf diseases significantly reduce both the quantity and quality of cassava production. To address this challenge, the study developed two enhanced machine learning models—Enhanced Binary Particle Swarm Optimisation-Support Vector Machine (EBPSO-SVM) and Enhanced Reptile Search Algorithm-Support Vector Machine (ERSA-SVM)—for improved multiclass classification of cassava leaf diseases. The research utilised a dataset of 2,180 cassava leaf images from the Kaggle village repository, comprising 466 images each of cassava bacterial blight disease (CBBD), cassava brown streak disease (CBSD), cassava green mottle or mite disease (CGMD), cassava mosaic disease (CMD), and 316 images of healthy leaves. Preprocessing involved image resizing, RGB to grayscale conversion, and contrast enhancement using bi-histogram equalisation. The affected areas were segmented using the Sobel edge detection method, while Gray Level Spatial Dependence and colour moment techniques were employed for texture, shape, and colour feature extraction. Comparative experiments revealed that both the EBPSO-SVM and ERSA-SVM models achieved superior performance with an average classification accuracy of 96.42% and 95.14%, respectively, outperforming both the BPSO-SVM and RSA-SVM models, which attained an average accuracy of 95.40% and 93.86%, respectively. These findings demonstrate the effectiveness of the enhanced optimisation algorithms in overcoming dataset imbalance, premature convergence, and local optima challenges that often hinder model accuracy. In conclusion, the proposed EBPSO-SVM and ERSA-SVM models enhance classification precision and efficiency in cassava disease detection. Their adoption could facilitate early and automated identification of cassava leaf diseases, contributing to improved crop management, increased agricultural productivity, and enhanced food security in cassava-dependent regions.