Classification of Segmented Milkfish Eyes using Cosine K-Nearest Neighbor

Prasetyo, Eko and Adityo, R. Dimas and Purbaningtyas, Rani (2019) Classification of Segmented Milkfish Eyes using Cosine K-Nearest Neighbor. In: 2019 2nd International Conference on Applied Information Technology and Innovation (ICAITI). IEEE Xplore . IEEE, Bali, Indonesia, pp. 93-98. ISBN 978-1-7281-3018-7

	Text (Artikel Classification segmented milk eyes) ICAITI 2019 - Classification of segmented milkfish eyes.pdf - Published Version Download (1MB)
	Text (Hasil similarity artikel classification milkfish eyes) Similarity - Classification of Segmented Milkfish Eyes Using Cosine K-Nearest Neighbor.pdf - Supplemental Material Download (1MB)
	Text (Hasil Peer Review) 15. Claification Of Segmented Milkfish Eyes using Cosin.pdf - Supplemental Material Restricted to Repository staff only Download (436kB)

Abstract

The classification of milkfish freshness based on the eyes is supported by the results of correct fish eye segmentation. Generally, the problems faced by segmentation results are many objects with similar characters and similar gray intensity. Segmentation with K-Means produces K layers of binary image according to the selected K cluster. As a result, many objects appear as the result of segmentation. Among all objects appear, the fish eye has a special character, where it is round black. The problem faced by classic K-Nearest Neighbor (KNN) in classification is sensitive to noise when using low K while using high K the classification performance falls to the most class that shouldn’t as the result. We propose Cosine KNN (CosKNN) to solve the classic KNN problem where the classification results aren’t taken from the most class of nearest neighbor. CosKNN gives soft value that represents the belonging level of each class to the testing data. To evaluate the performance of CosKNN, we use precision and recall. The experiment result shows that the CosKNN achieves performance both precision and recall of 97.93% and 91.15%, respectively, all with shape features. Especially on precision performance, CosKNN achieves the highest performance compared to other methods, CosKNN, classic KNN and K-SVNN achieve 97.93%, 96.20%, and 95.77%. While in recall performance, KNN achieves the highest performance compared to other methods, CosKNN, classic KNN and K-SVNN achieve 91.15%, 92.10%, and 90.76%, respectively, all with shape features.

Actions (login required)

View Item