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

Abstract

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.

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

“…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].…”

Mind the Traps! Design Guidelines for Rigorous BCI Experiments

“…Generalizing is different from memorizing the data (a.k.a overfitting): more details on this point are provided in the next section. Recent nonlinear methods that are somewhat robust against overfitting are Random Forest (Steyrl 2016) and neural networks such as Restricted Boltzmann Machines (Kober 2016). These methods do not make any assumption about the data distribution, and as such, are likely to be increasingly used for BCI in the future, and to give potentially better performances.…”

Personal Identification Based on Brain Networks of EEG Signals

“…Motor data. The Motor data is a public dataset obtained from the BNCI Horizon 2020 project. It is a two-class motor imagery task data set recorded by Steyrl et al (2016). There were 14 participants in the experiment.…”