Hybrid principal component analysis and K-nearest neighbour to detect the catfish disease

Catfish cultivation in Indonesia is a very promising business opportunity with a big profits. Every year, market demand continues to increase. However, this is contrary with the lack of catfish farmer's knowledge, so that catfish yields are not optimal. This is because, for certain types of catfish such as Sangkuriang catfish, it is easy to contract certain diseases. This study aims to create an automated system that capable of detecting catfish disease based on its symptoms with image recognition techniques. Early detection of catfish disease can help to find out the causes and prevention, so that the yield remains optimally. The method that used in this study is Principal Component Analysis (PCA) for feature extraction in images combined with K-Nearest Neighbour (KNN) with Euclidean Distance to classify catfish diseases among others: white spots, edema (abdominal swelling), jaundice, and bent spinal disease (scoliosis and lordosis). Based on the results of the experiment using 30 images data for training and 20 images data for testing, 18 image data is classified correctly. This result proves that PCA and KNN able to detect catfish disease well with percentage of accuracy around 90%.