OptNet: Optimized Tabnet For Lung Cancer Prediction
IOSR Journal of Mobile Computing & Application · 2026-03 · 13 (3) · 15-30
DOI: 10.9790/0050-13031530 ↗Kurz gefasst
Fachartikel in der Zeitschrift IOSR Journal of Mobile Computing & Application, 2026-03. Verfasst von A. Karunamurthy A. Karunamurthy und 1 weiteren Autor:innen.
Abstract (Originalsprache)
Lung cancer remains a leading cause of cancer-related mortality worldwide, and accurate predictive models from structured clinical data are essential for early diagnosis and treatment planning. We propose OptNet, a novel framework that integrates an attention-based TabNet classifier with automated hyperparameter optimization for high-performance lung cancer prediction. The core of our method is a sequential attention mechanism that performs instance-wise feature selection through sparse masking, thereby focusing computational resources on the most clinically relevant attributes—such as tumor size or biomarker levels— while suppressing noise from irrelevant variables. This design departs from conventional deep learning models that process all features uniformly. To further enhance predictive accuracy, we embed the Optuna optimization engine, which employs a Tree-structured Parzen Estimator algorithm to dynamically search the hyperparameter space. The objective function is the negative validation accuracy, and the optimizer iteratively samples new configurations—including the number of decision steps, feature transformer width, momentum, and sparsity coefficient—based on posterior distributions from previous trials. Each sampled configuration is immediately substituted into the TabNet architecture before training, creating a closed-loop refinement process that converges to a globally optimized model. The final output is a malignancy probability score for each patient. We evaluate OptNet on a structured lung cancer dataset comprising demographic, symptomatic, and imaging-derived features. Our results demonstrate that the integration of sparse attention with Bayesian hyperparameter search yields superior classification performance compared to standard baselines, while maintaining interpretability through the learned feature masks. The framework thus offers a robust and computationally efficient solution for clinical decision support in oncology.
Zitieren
APA
Karunamurthy, A. K. A., & Gayathiri, R. G. R. (2026). OptNet: Optimized Tabnet For Lung Cancer Prediction. IOSR Journal of Mobile Computing & Application, 13(3), 15-30. https://doi.org/10.9790/0050-13031530
Vancouver
Karunamurthy AKA, Gayathiri RGR. OptNet: Optimized Tabnet For Lung Cancer Prediction. IOSR Journal of Mobile Computing & Application. 2026;13(3):15-30. doi:10.9790/0050-13031530
BibTeX
@article{Karunamurthy2026,
author = {Karunamurthy, A. Karunamurthy A. and Gayathiri, R. Gayathiri R.},
title = {OptNet: Optimized Tabnet For Lung Cancer Prediction},
journal = {IOSR Journal of Mobile Computing & Application},
year = {2026},
volume = {13},
number = {3},
pages = {15--30},
doi = {10.9790/0050-13031530}
}
EUVIMED