Development of a Hybrid Machine Learning Model for Apple (Malus domestica) Health Detection and Disease Classification

Abstract

Apple is one of the top agricultural products produced each year. Even with apple one of the most cultivated crops, its demand is also increasing. As a result, this crop which is typically grown in temperate climate regions is now being cultivated in tropical climate regions. A factor affecting the production of apples each year is the pest and disease infestation. This study aims to create a machine learning model that can detect and classify the three most common 3 types of apple disease: apple scab (Venturia inaequalis), black rot (Botryosphaeria obtusa), and cedar apple rust (Gymnosporangium juniperi-virginianae). Color and texture feature extractions were performed to 400 single-apple leaf images followed by feature selection reducing the number of predictors to R, V, and b*. A comparison among K-Nearest Neighbors (KNN), Artificial Neural Network (ANN) and Gaussian Process Regression (GPR) was done to determine which is the best performing model that detects whether an apple leaf is healthy or infested. The GPR model with ARDSquared kernel function was determined to be the best performing model with training accuracy of 88.44%, testing accuracy of 82.50%, and <tex xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">$\text{AUC} =0.9256$</tex> . On the other hand, Decision Tree and Support Vector Machine (SVM) were compared to get the better performing model for classification disease classification. The quadratic SVM model was determined to be the better performing model obtaining 77.1% training accuracy, 83.3% testing accuracy, and <tex xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">$\text{AUC}=0.86$</tex> .