Skripsi

ANALISIS HYBRID FEATURE ENGINEERING UNTUK PENENTUAN TINGKAT KEPARAHAN LESI PRA-KANKER SERVIKS

---

Visual diagnosis via colposcopy is prone to observer subjectivity, making a more objective computational system necessary. This study explores two approaches: hybrid feature engineering (color, texture, contour) using machine learning (ML) via a rule-based system that adapts the Sweden score method, and end-to-end architectures based on YOLO (v8, v11, v12, v26). The dataset is sourced from the International Agency for Research on Cancer (IARC) and was independently split based on case status. Experiments were designed under two scenarios: three-class classification and binary classification. Preliminary results indicate chromatic parameters as the most discriminative features, achieving 67% accuracy via the XGBoost algorithm. In the final comparison, the performance of both approaches was relatively comparable in the three-class scenario (51%), but the Sweden score based ML approach significantly outperformed in binary classification (72.3%), surpassing the best performance of the YOLO variants (64.86%). These findings demonstrate that massive computational models like YOLO are prone to overfitting on limited medical data, while the combination of manual features and adapted clinical methods offers more robust generalization and interpretability relevant to medical computational analysis.

---

Availability

Detail Information

Series Title

-

Call Number

T2000382026

Publisher

Indralaya : Prodi Sistem Komputer, Fakultas Ilmu Komputer Universitas Sriwijaya., 2026

Collation

xvi, 125 hlm.; ilus.; tab.; 29 cm.

Language

Indonesia

ISBN/ISSN

-

Classification

616.994 07

Content Type

Text

Media Type

-

Carrier Type

-

Edition

-

Subject(s)

Prodi Sistem Komputer
Penyakit Kanker dan Tumor

Specific Detail Info

-

Statement of Responsibility

MI

Other version/related

File Attachment

Comments

---

You must be logged in to post a comment