You can find our paper here: TS-CATMA: A Lung Cancer Electronic Nose Data Classification Method Based on Adversarial Training and Multi-scale Attention.
conda create -n tscatma python=3.8.18
cd TS-CATMA
pip install -r ./requirements.txt
- Set the dataset path in
./data_loader.py. - Use the default configuration file or provide your own in
./configs/ - Run the following command to start training:
The results will be saved in the
python main.py./output/directory. - (Optional) You can set the path to a trained model in each
./plot*.pyscript to generate various visualization results.
If you use this work in your research, please cite the following paper:
@InProceedings{10.1007/978-981-96-0119-6_7,
author="Chen, Yuze
and Yi, Lin
and Wang, Shidan
and Tian, Fengchun
and Liu, Ran",
editor="Hadfi, Rafik
and Anthony, Patricia
and Sharma, Alok
and Ito, Takayuki
and Bai, Quan",
title="TS-CATMA: A Lung Cancer Electronic Nose Data Classification Method Based on Adversarial Training and Multi-scale Attention",
booktitle="PRICAI 2024: Trends in Artificial Intelligence",
year="2025",
publisher="Springer Nature Singapore",
address="Singapore",
pages="73--78",
abstract="Accurate lung cancer diagnosis is crucial for effective treatment and improved outcomes. This study introduces TS-CATMA (Time Series Classification with Adversarial Training and Multi-scale Attention), a novel method designed for lung cancer detection using electronic nose data. TS-CATMA leverages a multi-scale attention mechanism and adversarial training to extract discriminative, domain-invariant features from raw time series data. Evaluated on a lung cancer electronic nose dataset, TS-CATMA achieved a detection accuracy of 90.59{\%} with rapid training (6.15 s) and testing (39.57 ms) times, indicating its potential for early diagnosis. The source code is available at https://github.com/CQU-3DTEAM/TS-CATMA.",
isbn="978-981-96-0119-6"
}