Skip to content
SDGs Center UDINUS

Evidence record · UDINUS · SDGs

Effectiveness of K-means Clustering-Based Outlier Handling Techniques in Colonic Condition Endoscopy Image Classification Using CNN

Rektorat UDINUS 2024 Published 2026-10-30

Evidence record · UDINUS · SDGs

Unit
Rektorat UDINUS
Period
2024
Goals
Record
1569afa4-754b-4e05-ab9f-e7563eb15c47
Published
2026-10-30

Universitas Dian Nuswantoro (UDINUS), through researchers Athiyah, Ummi; Rustad, Supriadi; Soeleman, Moch Arief; Muljono; and Akrom, Muhamad, conducted a study on K-means Clustering-based outlier handling in the classification of colon endoscopic images using a convolutional neural network. The study addressed the need for accurate early detection in the context of colorectal cancer, identified in the source as one of the leading causes of cancer death worldwide.

The activity used endoscopic imaging because this technology plays a vital role in diagnosis. It examined the effect of outliers in endoscopic images, which can significantly affect the accuracy of a classification model. Histogram of Oriented Gradients was used for feature extraction. K-means Clustering was applied for outlier detection and removal. The dataset consisted of colon endoscopic images obtained from the public database Gastrovision. The images were divided into two classes: normal and abnormal conditions.

The modeling process consisted of two steps. The first step used the original dataset. The second step used the cleaned dataset after outlier removal. The reported results showed improved model performance after outlier handling. Accuracy increased from 61% to 85%. The Area Under Curve of the ROC curve increased from 0.84 to 0.92. The increase in AUC indicated a better model ability to distinguish between normal and abnormal conditions.

The findings emphasize the importance of data quality and effective outlier management in training a machine-learning model for colon cancer endoscopic image conditions. The activity therefore frames UDINUS participation in research that applies feature extraction, outlier detection and removal, and image classification to endoscopic images. The reported comparison between the original and cleaned datasets provides the stated basis for assessing model performance after outlier handling.

Reference https://www.scopus.com/pages/publications/85218491713?origin=resultslist