Skip to content
SDGs Center UDINUS

Evidence record · UDINUS · SDGs

Multi-Strategy Feature Selection for Glioma Grading Using Classical and Deep Models

Rektorat UDINUS 2025 Published 2026-10-30

Evidence record · UDINUS · SDGs

Unit
Rektorat UDINUS
Period
2025
Goals
Record
f9d17050-f7fd-4fd3-b561-467a14be844d
Published
2026-10-30

Universitas Dian Nuswantoro conducted an activity on glioma grading through an adaptive glioma classification pipeline. The activity involved UDINUS researchers Rizqi, Ainur Rahma Miftakhul; Nugraha, Adhitya; Luthfiarta, Ardytha; Dewi, Ika Novita; Hafiizhudin, Lutfi Azis; and Novtagravina, Adelfia Milka. Glioma was addressed as a complex brain tumor for which accurate grading is essential for timely effective treatment planning and diagnosis. The activity focused on challenges involving high-dimensional clinical and molecular features and significant class imbalance, which can hinder model performance.

The evaluation used a subset of The Cancer Genome Atlas dataset comprising 839 patient records with 24 features. Four classification models were evaluated: Logistic Regression, Random Forest, Multilayer Perceptron, and CNN-1D. The models were combined with three feature selection methods: Mutual Information, Particle Swarm Optimization, and Hybrid Voting. Performance was assessed using both imbalanced data and data balanced through SMOTE. This design enabled comparison of classification models, feature selection strategies, and data conditions within the evaluated pipeline.

The most robust performance was achieved by CNN-1D with PSO-based feature selection. This combination yielded an accuracy of 88.1%, an F1-score of 88.2%, an AUC of 92.0%, a precision of 88.9%, and a recall of 88.1%. The application of SMOTE provided a statistically significant improvement for the Random Forest model, with p < 0.005. SMOTE also enabled the Logistic Regression model to achieve a peak AUC of 93.0% and a recall of 94.3%.

The findings confirm that integrating data balancing with adaptive feature selection can improve glioma classification performance. The resulting approach was presented as a reliable decision support tool for medical professionals. The activity therefore frames UDINUS research through the evaluation of computational methods for glioma classification and their stated relevance to diagnosis and treatment planning.

Reference https://www.scopus.com/pages/publications/105031617735?origin=resultslist