EEG Datasets¶

EEG Data Characteristics¶

Problem	train cases	test cases	dimensions	length	num classes
Blink	500	450	4	510	2
EyesOpenShut	56	42	14	128	2
FaceDetection	5890	3524	144	62	2
FingerMovements	316	100	28	50	2
HandMovementDirection	160	74	10	400	4
MindReading	727	653	204	200	2
MotorImagery	278	100	64	3000	2
SelfRegulationSCP1	268	293	6	869	2
SelfRegulationSCP2	200	180	7	1152	2

Classification¶

Benchmark classification experiments, results generated with code like this

        dataset = "EyesOpenShut"
        results_dir = "C:/temp/" # where to results
        data_dir = "C:/temp/" # Location of data in ts format data_dir/<dataset>/
        resample = 0        # 0 indicates default train test
        cls_name = "RocketClassifier" # name used for results name
        from aeon.classification.kernel_based import RocketClassifier
        from aeon.benchmarking.experiments import \
            load_and_run_classification_experiment
        classifier = RocketClassifier()
        load_and_run_classification_experiment(
            overwrite=False,
            problem_path=data_dir,
            results_path=results_dir,
            cls_name=cls_name,
            classifier=classifier,
            dataset=dataset,
            resample_id=resample,
            build_train=False,
            predefined_resample=False,
        )

This generates a result file called testResample0.csv. Some exploratory results:

Best classifier accuracies (default train split, default classifier settings)¶

Problem	Best Acc	Best Classifier
Blink	1.000000	Arsenal/RocketClassifier
EyesOpenShut	0.523810	DrCIF/MUSE/RocketClassifier
FaceDetection	0.678490	CNNClassifier
FingerMovements	0.560000	ShapeletTransformClassifier
HandMovementDirection	0.581081	CNNClassifier
MindReading	0.595712	FreshPRINCE
MotorImagery	0.590000	HIVECOTEV2
SelfRegulationSCP1	0.897611	FreshPRINCE
SelfRegulationSCP2	0.538889	KNeighborsTimeSeriesClassifier/TDE

Best aeon classifier accuracies (30 resamples, default classifier settings)¶

Problem	Best Acc	Best Classifier
Blink	0.9999	TemporalDictionaryEnsemble
EyesOpenShut	0.6643	KNeighborsTimeSeriesClassifier
FaceDetection	0.7187	CNNClassifier
FingerMovements	0.5813	Mini-ROCKET
HandMovementDirection	0.5775	CNNClassifier
MindReading	0.7369	Mini-ROCKET
MotorImagery	0.5423	TemporalDictionaryEnsemble
SelfRegulationSCP1	0.9110	Multi-ROCKET
SelfRegulationSCP2	0.5531	MUSE

Results discussion¶

Blink

EyesOpenShut

FaceDetection (MEG) The [leaderboard](https://www.kaggle. com/competitions/decoding-the-human-brain/leaderboard) shows a best accuracy of 0. 75501, although it is not clear if the results are directly comparable to those above. This needs clarification.

FingerMovements

HandMovementDirection