This paper introduces a novel approach to identifying and classifying herbal plant species using leaf images. While deep learning models have shown remarkable performance in this domain, their decision-making process often lacks transparency, making it challenging to understand the reasoning behind their classifications. The proposed model is applied specifically to the classification of leaf images, providing significant insights into the characteristics of herbal plant species. Unlike existing models, the proposed model uniquely leverages a variational auto-encoder (VAE) to simultaneously achieve high classification accuracy and offer feature-based explanations. The VAE architecture enhances explainability by disentangling classifiable features from non-classifiable ones, allowing for clear visualization of the most important features driving classification decisions. This aspect has not been fully explored in previous works. Our approach demonstrated superior classification accuracy of 99.20% and enhanced explainability through experiments conducted on the BDMediLeaves dataset. Our model enabled visual understanding of important features influencing classification decisions, empowering researchers with actionable insights into botanical image classification. The explainable nature of our model enhances transparency and trust in automated plant species identification systems, crucial for validation and knowledge extraction. This model shows significant potential for broader applications in other crops and diverse image classification tasks, ensuring both transparency and high performance.