An Explainable AI Approach for Potato Plant Disease Detection using Enhanced Feature Engineering

Agriculture plays a significant role in India’s fast-growing economy. It makes up a substantial portion of GDP and employs a sizable section. Among the various crops, potato cultivation is one of the most significant practices since it contributes to food security and is a cash crop for most farmers. However, there are number of illnesses that can negatively affect productivity and quality of potato. There have been several advancements made to address these issues through machine learning techniques. This study presents detection of potato plant diseases by image processing of leaves plagued by the various potato diseases. The application of feature extraction methods such as Hough Transform (HT) and Discrete Wavelet Transform (DWT) were explored across various color spaces i.e. RGB, YCbCr, HSV and LaB, followed by the implementation of multiple ML classifiers such as Ada Boost (AB), Random Forest (RF), Naïve Bayes (NB), Support Vector Machine (SVM) and Gradient Boosting (GB) and logistic regression, on the extracted features. The study shows that logistic regression has achieved the highest accuracy of 0.99 in YCbCr color space with 256 bins. To enhance the knowledge about feature contribution, feature importance analysis was done, and it was found that DWT significantly contributed for efficient model performance. This promoted the XAI approach explained here.