The agriculture includes tea as an important component. The leaves of the tea plant are infected with various diseases that can affect both the crop and the environment. Disease detection at the mild stage helps minimize crop losses. A comprehensive tea leaf dataset was created, consisting of 9000 high-quality healthy and diseased leaf images. The dataset includes five diseased classes: Blister Blight, Red Spider Mite, Brown Blight, Leaf Red Rust, and Tea Mosquito Bug. There are 1,500 images per class, making the dataset class-balanced. Images are pre-processed using resize, flip, zoom, shift, shear, rotate, and median filter. After preprocessing, each class contains 13,500 images. The threshold-based segmentation techniques, such as Lab-Otsu and HSV + ExG, were then applied to 81,000 images, yielding segmented masks for both. Based on the applied rules, the best mask is selected. The segmented images are further extracted using the masks. We then extract features, including color, GLCM texture, and shape, from the segmented images. Lastly, various machine learning methods, including Gradient Boosting, XGBoost, RF, Extra Tree, K-NN, and DT, utilize the features to predict the tea leaf image category. Among all approaches, the XGBoost classifier achieved the highest value of 97%.