Team name: MIWEAR
Aims: The primary goal of this study is to enhance the prognostic accuracy of electroencephalography (EEG) data for patients after cardiac arrest by employing machine learning and deep learning algorithms, ultimately improving patient outcomes. Methods: We utilized a dataset of 607 patients and employed stratified K-Fold cross-validation (k=2, stratified on the CPC score) for evaluation. By transforming the Cerebral Performance Category (CPC) scale into a binary "outcome" class, we treated the problem as a regression task. Our base model, the LightGBM regressor, leverages demographic features such as age, sex, and medical history, as well as statistical features (mean, standard deviation, minimum, maximum) of EEG data for prediction. The model's performance on different features and their combinations was analyzed to identify key variables and optimize the model. Results: The cross-validation challenge score was 0.31 (average of fold scores: 0.41 and 0.20), and the leaderboard challenge score was 0.33. Despite the modest scores, our focus at this stage is on laying a solid foundation for further development and refining the methodology, exploring the potential of various machine learning and deep learning techniques in this domain. Future Work: We plan to incorporate additional features such as reduced voltage, burst suppression, seizures, and seizure-like patterns. Furthermore, we will explore the application of convolutional neural networks and other advanced techniques to further enhance the prognostic accuracy of our model in post-cardiac arrest care.