Random Forest and Filter Bank Common Spatial Patterns for EEG-Based Motor Imagery Classification

Bentlemsan, Maouia; Zemouri, ET-Tahir; Bouchaffra, Djamel; Yahya-Zoubir, Bahia; Ferroudji, Karim

doi:10.1109/isms.2014.46

Cited by 48 publications

(29 citation statements)

References 19 publications

Supporting

Mentioning

Contrasting

Unclassified

Order By: Relevance

“…For classification, we use Random Forests (RF) classifiers [34] which are resistant to over-fitting even at low trials-to-features ratio, robust to outliers, and capable to build a complex model for capturing non-linear relationships while running in (near) real-time. RF classifiers have been recently applied successfully in decoding EEG signals for both classification [35]- [37] and data mining purposes [38] as well as in other fields of science such as data mining [39], gene analysis [40] and computer vision [41]- [43].…”

Section: Methodsmentioning

confidence: 99%

A brain-controlled exoskeleton with cascaded event-related desynchronization classifiers

Lee

Liu

Perroud

et al. 2017

Robotics and Autonomous Systems

116

View full text Add to dashboard Cite

This paper describes a brain-machine interface for the online control of a powered lower-limb exoskeleton based on electroencephalogram (EEG) signals recorded over the user's sensorimotor cortical areas. We train a binary decoder that can distinguish two different mental states, which is applied in a cascaded manner to efficiently control the exoskeleton in three different directions: walk front, turn left and turn right. This is realized by first classifying the user's intention to walk front or change the direction. If the user decides to change the direction, a subsequent classification is performed to decide turn left or right. The user's mental command is conditionally executed considering the possibility of obstacle collision. All five subjects were able to successfully complete the 3-way navigation task using brain signals while mounted in the exoskeleton. We observed on average 10.2% decrease in overall task completion time compared to the baseline protocol.

show abstract

Section: Methodsmentioning

confidence: 99%

A brain-controlled exoskeleton with cascaded event-related desynchronization classifiers

Lee

Liu

Perroud

et al. 2017

Robotics and Autonomous Systems

116

View full text Add to dashboard Cite

show abstract

“…In this study, we show the feasibility of applying the random forest algorithm [24] to single-trial source-localized time-frequency data. With its non-parametric, non-linear approach and its capability to handle high dimensional datasets with highly correlated predictors, this method is well suited for MEG data (also see [33,34,35,36]) and can detect subtle differences concealed in the averaged data.…”

Section: Discussionmentioning

confidence: 99%

“…Gamma band activity was estimated using frequency smoothing (Slepian sequence multi taper approach), yielding 20 Hz-wide frequency bands centered at 35,65,85,115 and 135 Hz. The power was calculated separately for 250 ms long time windows from −500 to 1000 ms and the post-stimulus activity was then expressed as relative change to baseline power.…”

Section: Source Reconstructionmentioning

confidence: 99%

Across-subjects classification of stimulus modality from human MEG high frequency activity

Westner

Dalal

Hanslmayr

et al. 2017

Preprint

View full text Add to dashboard Cite

Single-trial analyses have the potential to uncover meaningful brain dynamics that are obscured when averaging across trials. However, low signal-to-noise ratio (SNR) can impede the use of single-trial analyses and decoding methods. In this study, we investigate the applicability of a single-trial approach to decode stimulus modality from magnetoencephalography (MEG) high frequency activity. In order to classify the auditory versus visual presentation of words, we combine beamformer source reconstruction with the random forest classification method. To enable group level inference, the classification is embedded in an across-subjects framework.We show that single-trial gamma SNR allows for good classification performance (accuracy across subjects: 66.44 %). This implies that the characteristics of high frequency activity have a high consistency across trials and subjects. The random forest classifier assigned informational value to activity in both auditory and visual cortex with high spatial specificity. Across time, gamma power was most informative during stimulus presentation. Among all frequency bands, the 75-95 Hz band was the most informative frequency band in visual as well as in auditory areas. Especially in visual areas, a broad range of gamma frequencies (55-125 Hz) contributed to the successful classification.Thus, we demonstrate the feasibility of single-trial approaches for decoding the stimulus modality across subjects from high frequency activity and describe the discriminative gamma activity in time, frequency, and space. Author SummaryAveraging brain activity across trials is a powerful way to increase signal-to-noise ratio in MEG data. This approach, however, potentially obscures meaningful brain dynamics that unfold on the single-trial level. Single-trial analyses have been successfully applied to time domain or low frequency oscillatory activity; its application to MEG high frequency activity is hindered by the low amplitude of these signals. In the present study, we show that stimulus modality (visual versus auditory presentation of words) 2 not peer-reviewed) is the author/funder. All rights reserved. No reuse allowed without permission.The copyright holder for this preprint (which was . http://dx.doi.org/10.1101/202424 doi: bioRxiv preprint first posted online Oct. 12, 2017; can successfully be decoded from single-trial MEG high frequency activity by combining source reconstruction with a random forest classification algorithm. This approach reveals patterns of activity above 75 Hz in both visual and auditory cortex, highlighting the importance of high frequency activity for the processing of domain-specific stimuli. Thereby, our results extend prior findings by revealing high-frequency activity in auditory cortex related to auditory word stimuli in MEG data. The adopted across-subjects framework furthermore suggests a high inter-individual consistency in the high frequency activity patterns.

show abstract

“…This weighting improves separability, and therefore, in general, also accuracy of SMR-BCIs. Although fi rst works on the combination of CSP fi ltering and RF classifi ers are available [2,5] , a direct comparison with the common LDA classifi er on CSP fi ltered features is missing. We performed a BCI simulation with the data of the online SMR-BCI to provide such a comparison.…”

Section: Bci Simulation: Random Forest Vs Regularized Ldamentioning

confidence: 99%

Random forests in non-invasive sensorimotor rhythm brain-computer interfaces: a practical and convenient non-linear classifier

Steyrl

Scherer

Faller

et al. 2016

Biomedical Engineering / Biomedizinische Technik

View full text Add to dashboard Cite

There is general agreement in the brain-computer interface (BCI) community that although non-linear classifiers can provide better results in some cases, linear classifiers are preferable. Particularly, as non-linear classifiers often involve a number of parameters that must be carefully chosen. However, new non-linear classifiers were developed over the last decade. One of them is the random forest (RF) classifier. Although popular in other fields of science, RFs are not common in BCI research. In this work, we address three open questions regarding RFs in sensorimotor rhythm (SMR) BCIs: parametrization, online applicability, and performance compared to regularized linear discriminant analysis (LDA). We found that the performance of RF is constant over a large range of parameter values. We demonstrate - for the first time - that RFs are applicable online in SMR-BCIs. Further, we show in an offline BCI simulation that RFs statistically significantly outperform regularized LDA by about 3%. These results confirm that RFs are practical and convenient non-linear classifiers for SMR-BCIs. Taking into account further properties of RFs, such as independence from feature distributions, maximum margin behavior, multiclass and advanced data mining capabilities, we argue that RFs should be taken into consideration for future BCIs.

show abstract

Random Forest and Filter Bank Common Spatial Patterns for EEG-Based Motor Imagery Classification

Cited by 48 publications

References 19 publications

A brain-controlled exoskeleton with cascaded event-related desynchronization classifiers

A brain-controlled exoskeleton with cascaded event-related desynchronization classifiers

Across-subjects classification of stimulus modality from human MEG high frequency activity

Random forests in non-invasive sensorimotor rhythm brain-computer interfaces: a practical and convenient non-linear classifier

Contact Info

Product

Resources

About