Investigating a Spectrum of Machine Learning Methods for Leaf Disease Detection in Pepper, Potato, and Tomato

An analysis of machine learning (ML) techniques such as K-Nearest Neighbor (KNN), Support Vector Machine (SVM), Random Forest (RF) and Naive Bayes (NB) was done to detect diseases among pepper, potato, and tomato crops to develop an algorithm to extract multiple features such as kurtosis, GLCM energy, skewness etc and to find region of interest (ROI). First, the sample images were resized and converted into greyscale. Then k-means clustering was used for segmentation. The cluster with the highest contrast or most significant visual features was chosen as the ROI and then passed to the machine learning models to predict the accuracy of finding the diseases. For experimentation purpose 20,639 leaf images of potato, tomato, and pepper plant is taken. From the dataset, 70% were used for training, while 30% of these images were used for testing. The data (biased and unbiased) was passed to these models in shuffled and un-shuffled orders, and it was found that Random Forest gave the highest accuracy i.e. 94.2% in the case of biased data and 89.90% with unbiased data. Hence, it is observed that the proposed algorithm works well for large number of images.