The data can be downloaden from Physionet. Moreover, a simple script to download the data is provided: download_data.sh.
The structure of the directory should be: data/tpehgdb/tpehg<ID>.{dat,hea}
To install the required dependencies, run pip install -e .. To install our package, run (sudo) python3 setup.py install
Features are grouped per study. To extract features (from multiple studies), the following construct can be used:
from ehgfeatures.features import FeatureGroup, FeaturesAcharya, FeaturesHosseinzahde
from ehgfeatures.signal_io import get_signals
DATA_PATH = 'data/tpehgdb'
ids, signals, gestations, remaining_durations = get_signals(DATA_PATH, n_signals=1)
signal = signals[0][0]
fe = FeatureGroup([FeaturesAcharya(), FeaturesHosseinzahde()])
results = fe.extract(signal)We generate uniformly distributed data and fit a Random Forest on data, oversampled with SMOTE before partitioning in train & test and after partitioning: python3 experiments/smote_random_data.py
AUC no oversampling: 0.49211624313186814
AUC with oversampling after partitioning: 0.48493389423076927
AUC with oversampling before partitioning: 0.9546904407849975
First, extract all the features using our provided script: python3 experiments/all_features.py to generate a file with features for each signal and channel, followed by python3 experiments/process_feature_files.py to create a raw_features.csv file with the features of the different channels joined for each signal. Finally, run python3 generate_auc.py:
Channel 1 (all) Channel 1 (early) Channel 1 (late)
mean std mean std mean std
Feature
FeaturesAcharya_aaa_emd_1_FeatureSampleEntropy 0.591544 0.051503 0.480708 0.095833 0.652101 0.043442
FeaturesAcharya_aaaa_emd_1_FeatureStandardDevia... 0.401349 0.032703 0.487821 0.084913 0.326017 0.039307
FeaturesAcharya_aaaa_emd_1_FeatureTeagerKaiserE... 0.420838 0.050610 0.460950 0.060132 0.321763 0.054654
FeaturesAcharya_aad_emd_1_FeatureInterquartileR... 0.568805 0.051237 0.624491 0.069304 0.583167 0.048826
FeaturesAcharya_aad_emd_6_FeatureFractalDimensi... 0.640088 0.058785 0.715739 0.062912 0.618584 0.086977
FeaturesAhmedsampen_4_5 0.444810 0.040487 0.451224 0.077012 0.503737 0.085226
FeaturesHosseinzahde_aaa_emd_1_FeatureAR_Yule_W... 0.361289 0.060995 0.416381 0.058634 0.333392 0.047579
FeaturesJanjarasjitt_d_4_-_3 0.447276 0.041506 0.436094 0.060452 0.423270 0.064175
FeaturesRen_0_frequency 0.382981 0.026988 0.362142 0.063680 0.427883 0.110927
FeaturesSadiAhmed_emd_6_fwl_peak_power 0.342523 0.042130 0.361544 0.057213 0.339339 0.085714
Channel 2 (all) Channel 2 (early) Channel 2 (late)
mean std mean std mean std
Feature
FeaturesAcharya_aaa_emd_1_FeatureSampleEntropy 0.564813 0.045268 0.575717 0.043638 0.564077 0.109719
FeaturesAcharya_aaaa_emd_1_FeatureStandardDevia... 0.464688 0.060027 0.506590 0.062906 0.379597 0.071247
FeaturesAcharya_aaaa_emd_1_FeatureTeagerKaiserE... 0.430969 0.035297 0.464417 0.078161 0.358472 0.054636
FeaturesAcharya_aad_emd_1_FeatureInterquartileR... 0.572280 0.050424 0.605744 0.066559 0.526313 0.050452
FeaturesAcharya_aad_emd_6_FeatureFractalDimensi... 0.648276 0.045577 0.593505 0.064385 0.631572 0.057885
FeaturesAhmedsampen_4_5 0.457637 0.035524 0.399258 0.075491 0.438623 0.063016
FeaturesHosseinzahde_aaa_emd_1_FeatureAR_Yule_W... 0.419926 0.047405 0.458098 0.083623 0.374575 0.085831
FeaturesJanjarasjitt_d_4_-_3 0.471179 0.021272 0.468563 0.054219 0.463482 0.072305
FeaturesRen_0_frequency 0.481301 0.052345 0.455075 0.073036 0.490442 0.081626
FeaturesSadiAhmed_emd_6_fwl_peak_power 0.440030 0.063030 0.446934 0.089578 0.392018 0.060133
Channel 3 (all) Channel 3 (early) Channel 3 (late)
mean std mean std mean std
Feature
FeaturesAcharya_aaa_emd_1_FeatureSampleEntropy 0.673919 0.048000 0.733307 0.051992 0.638636 0.024737
FeaturesAcharya_aaaa_emd_1_FeatureStandardDevia... 0.352152 0.060521 0.380530 0.043683 0.382764 0.061430
FeaturesAcharya_aaaa_emd_1_FeatureTeagerKaiserE... 0.347506 0.044559 0.340020 0.073617 0.336005 0.063156
FeaturesAcharya_aad_emd_1_FeatureInterquartileR... 0.637819 0.029157 0.653283 0.044005 0.610691 0.088115
FeaturesAcharya_aad_emd_6_FeatureFractalDimensi... 0.581420 0.045892 0.540443 0.074312 0.652712 0.076683
FeaturesAhmedsampen_4_5 0.331351 0.051295 0.269554 0.065582 0.429794 0.063348
FeaturesHosseinzahde_aaa_emd_1_FeatureAR_Yule_W... 0.332379 0.039977 0.283418 0.041726 0.391753 0.052815
FeaturesJanjarasjitt_d_4_-_3 0.358783 0.036714 0.273106 0.058694 0.449618 0.075412
FeaturesRen_0_frequency 0.382948 0.050449 0.341886 0.064592 0.426213 0.061082
FeaturesSadiAhmed_emd_6_fwl_peak_power 0.500180 0.057612 0.498116 0.110596 0.444011 0.071167
Make sure the features are extracted and joined using all_features.py and process_feature_files.py. Then, run python3 examples/studies.py
fergus2013 husain khan peng
in_sample_auc 0.998988 0.872483 0.872483 0.999447
incorrect_oversampling_auc 0.967071 0.855769 0.986538 0.917242
with_oversampling_auc 0.59747 0.238255 0.815436 0.468076
without_oversampling_auc 0.44332 0.875839 0.832215 0.568684