Comparative Analysis of Machine Learning Algorithms for Binary Classification of Tea Leaf Diseases

India stands as one of the world's largest tea exporters. This requires for an implementation of an efficient and robust system for disease detection and prevention. The most important challenge faced by the tea industry is tea leaf diseases due to continuous pathogen infestation. These diseases can cause full-size crop losses, adversely affecting the economic system and the livelihood of tea farmers. The ability to detect these diseases early is essential. Early detection can cause timely intervention that can notably reduce harm to tea plants and keep them at their best possible quality. Unfortunately, traditional diagnostic techniques often depend on visible inspection and are hence inefficient as those methods can be gradual, labour in-depth and vulnerable to human errors, reducing the effectiveness in larger fields. This research paper aims to classify whether a leaf is diseased or healthy through a two-step process involving feature extraction using a pretrained Convolutional Neural Network (CNN) and classification using four machine learning models - LightGBM, CatBoost, Autoencoders and Self Organized Maps (SOMs). This paper utilizes a pre-trained VGG16 model, which is trained on the ImageNet dataset. This model is capable of extracting high-level features from images.