Effect of stationarity on traditional machine learning models: Time series analysis

Dixit, Ankit; Jain, Shikha

doi:10.1145/3474124.3474167

Cited by 4 publications

(2 citation statements)

References 7 publications

Supporting

Mentioning

Contrasting

Order By: Relevance

“…The Hamming window was used to reduce spectral leakage. From the power spectral density (PSD), spectral band power was extracted in the alpha band (8)(9)(10)(11)(12)(13) and in the beta band (20)(21)(22)(23)(24)(25)(26)(27)(28)(29)(30).…”

Section: Comparing Erds Between Early and Late Sessionsmentioning

confidence: 99%

“…However, machine learning algorithms also assume that the relationship between the predictive features and the command of interest are consistent throughout the trials. If the mapping changes throughout the dataset, it could cause the machine learning model to perform poorly 21 . We argue that such nonstationarities could be present in BMI datasets due to neural efficiency, a phenomenon where the brain becomes more efficient in executing well-practiced skills.…”

Section: Introductionmentioning

confidence: 99%

See 1 more Smart Citation

Repetitive execution of a reach-and-lift task is associated with inter-trial changes in movement-related EEG features and their predictive power

Paek

Prashad

2023

Preprint

View full text Add to dashboard Cite

Brain Machine Interfaces (BMIs) are systems that can help restore motor function to individuals with paralysis. These systems utilize machine learning algorithms to detect how the user wants to control an assistive device from their brain activity. These algorithms require a training dataset to infer patterns between a user's brain waves and the control signal for the assistive device. These training datasets consist of brain activity recordings while participants perform a very repetitive motor task. We hypothesize that these training datasets can be compromised due to the high number of trials. We anticipate that the movement-related brain waves of interest attenuate over time due to neural efficiency, where the brain becomes more efficient with practice in a motor task. To explore this hypothesis, we used an open-access EEG dataset where participants performed a simple reach-and-lift task. From each trial, time windows associated with resting and movement periods were segmented. Alpha and beta-band spectral power was estimated for each of these epochs and event-related desynchronization (ERD) was estimated as the suppression in spectral power from rest to movement. These ERDs were compared between early and late trials in the dataset. We also used linear discriminant analysis to assess a machine learning algorithm's accuracy in classifying whether the time windows belonged to rest or movement based on the spectral power band. We found that earlier trials had stronger ERDs and larger classification accuracies, suggesting that the repetition in the motor task causes attenuation of movement-related brain wave patterns and reduces efficacy of the dataset. These results call for a reevaluation of BMI performance in datasets that have numerous trials and an exploration of strategies that can compensate for longitudinal changes in movement-related brain activity used for BMIs.

show abstract

Section: Comparing Erds Between Early and Late Sessionsmentioning

confidence: 99%

Section: Introductionmentioning

confidence: 99%

Repetitive execution of a reach-and-lift task is associated with inter-trial changes in movement-related EEG features and their predictive power

Paek

Prashad

2023

Preprint

View full text Add to dashboard Cite

show abstract

Exploring the Effect of Sequence Smoothness on Machine Learning Accuracy

Wang,

Quach

2024

Lecture Notes in Networks and Systems

View full text Add to dashboard Cite

Contemporary Approaches to Analyze Non-stationary Time-series: Some Solutions and Challenges

Dixit

Jain

2022

RACSC

View full text Add to dashboard Cite

Enhancement of technology yields more complex time-dependent outcomes for better understanding and analysis. These outcomes are generating more complex, unstable, and high-dimensional data from non-stationary environments. Hence, more challenges are arising day by day to fulfill the increasing demand for future estimation. Thus, in this paper, an extensive study has been presented to comprehend the statistical complexity and randomness of non-stationary time series (NS-TS) data at the atomic level. This survey briefly explains the basic principles and terms related to non-stationary time series (NS-TS). After understanding the fundamentals of NS-TS, this survey categorized non-stationarity into groups and their subgroups based on a change in statistical behavior. It is followed by a comprehensive discussion on contemporary approaches proposed by researchers in each category of non-stationarity. These algorithms include clustering, classification, and regression techniques to deal with different types of domains. Every category of non-stationarity consists of a separate table to draw some advantages and disadvantages of existing approaches. At the end of each non-stationarity type, a short discussion and critical analysis have been done. In the conclusion section, it observed that this research sphere still has so many open challenges that need to be addressed and demand more exploration. Furthermore, it discusses the possible solution of improvisation in future research.

show abstract

Effect of stationarity on traditional machine learning models: Time series analysis

Cited by 4 publications

References 7 publications

Repetitive execution of a reach-and-lift task is associated with inter-trial changes in movement-related EEG features and their predictive power

Repetitive execution of a reach-and-lift task is associated with inter-trial changes in movement-related EEG features and their predictive power

Exploring the Effect of Sequence Smoothness on Machine Learning Accuracy

Contemporary Approaches to Analyze Non-stationary Time-series: Some Solutions and Challenges

Contact Info

Product

Resources

About