Performance comparison of traditional machine learning and deep learning methods in plant disease detection

Traditional machine learning (TML) and deep learning (DL) are two important approaches in artificial intelligence. To build predictive models, TML uses algorithm such as support vector machines (SVMs), k-nearest neighbors and decision trees. In this approach, feature extraction is the most important process that affects the algorithm's performance. While TML achieves successful results in small datasets, it struggles to show the same success in complex data and large-scale datasets. DL, a subset of machine learning, uses multilayer neural networks to automatically learn hierarchical feature representations from raw data. While it achieves very good results in fields such as finance, industry, medicine, and in studies such as image classification, natural language processing, and voice recognition-processing, it requires a large amount of data and computational resources. This study presents a computer-aided diagnosis of plant diseases to compare the predictive performance of TML and DL. Plants are a source of food and life of great importance for humans and other living things. Detection of plant diseases is very important in modern agriculture to minimize product losses and ensure food safety. Early and accurate identification of plant diseases can reduce pesticide use, increase crop yields and improve product quality. For this purpose, the performance of traditional methods such as LR, SVM, and RF was compared DL methods such as convolutional neural network (CNN), LSTM, YOLOv8 and VGG16 algorithms. The results were obtained as an average accuracy of 77,9%, with the highest being 83% for SVM in TML, and an average accuracy of 91,8%, with the highest being 99% for CNN in DL.