Citation

Munzir, Mohammad and Khan, Mohammad Shadab and Kumaresan, Prabha and Lee, Fong Yee and Kamal, Shahid (2026) Traditional to Transformer TabPFN for ASD Detection. In: 2026 Second International Conference on Emerging Computational Intelligence (ICECI). Institute of Electrical and Electronics Engineers Inc., pp. 1-5. ISBN 979-831953337-1

Text 24.pdf - Published Version Restricted to Repository staff only Download (335kB)

Abstract

Autism Spectrum Disorder (ASD) is a neurodevelopmental condition that typically includes challenges in communication, eye contact, social interaction and repetitive behaviors. Machine Learning approaches whether it be traditional or modern have been widely used for structured autism datasets; however, they require extensive feature engineering and may not work properly with small or imbalanced data. The dataset was preprocessed, merged and cleaned to form a unified dataset of 5,584 samples. TabPFN, a transformer-based model was evaluated alongside state-of-the-art models XGBoost and CatBoost. The models were evaluated on accuracy, R2 score, Mean Square Error (MSE), and Mean Absolute Error (MAE). The results show that TabPFN performed the best with accuracy of 88% test accuracy and 0.52 of R2 score. The improvement compared to CatBoost and XGBoost in terms of accuracy is 2% and 15- 18% in error reduction. These findings indicate that TabPFN has strong generalization capabilities with minimum tuning required and hold high potential for ASD diagnosis. This work also show potential for deep learning architecture for ASD screening. Future work can focus on working on large dataset with use of multimodal approach.

Item Type:	Book Section
Uncontrolled Keywords:	Deep learning, machine learning
Subjects:	Q Science > QA Mathematics > QA71-90 Instruments and machines > QA75.5-76.95 Electronic computers. Computer science
Divisions:	Faculty of Computing and Informatics (FCI)
Depositing User:	Ms Rosnani Abd Wahab
Date Deposited:	02 Jul 2026 01:47
Last Modified:	02 Jul 2026 01:47
URII:	http://shdl.mmu.edu.my/id/eprint/16196

Downloads

Edit (login required)