Zahra Asiri / Mathematics / Faculty Mentor: Pedro Maia
This research addresses a pervasive challenge in the analysis of Local
Field Potentials (LFPs) derived from rodent brain recordings, namely, the confounding presence of outliers stemming from diverse sources such as power line interference, electrode noise, and atypical neural activity. The discernment between genuine outliers indicative of unusual brain signals and those arising from experimental artifacts is crucial for avoiding erroneous scientific inferences. Leveraging various unsupervised outlier detection techniques, this study introduces an innovative Cumulative Outlier
Score (COS) approach to distinguish between genuine and spurious outliers.
The research commences with comprehensive dataset preparation, in-
corporating the collaborative expertise of domain specialists. Exploratory
data analysis unveils variable distributions and identifies conspicuous out-
liers, setting the stage for subsequent unsupervised outlier detection. The
PyOD library is deployed, employing algorithms such as local outlier fac-
tor, cluster-based LOF, and isolation forest.
Crucially, the research hypothesizes that genuine outliers will yield
intermediate COS, while artifact outliers will exhibit higher scores. The
study further proposes a unique approach to validate this hypothesis by
analyzing data from distinct experimental groups, exploring the impact
of removing high and intermediate COS points on group separability.
Anticipated outcomes include the establishment of an outlier profile
for each data point, informed by multiple detection techniques, and the
computation of a cumulative outlier score. These results not only promise
a refined understanding of LFP-related variables susceptible to outlier
distortions but also offer valuable insights for future analyses in electro-
physiology. The potential success of the innovative COS approach could
significantly enhance the accuracy and reliability of differentiating genuine
outliers from artifacts, thereby advancing the field’s capacity to draw valid
scientific conclusions from LFP experiments.
This research not only contributes to the methodological arsenal of
outlier detection in electrophysiological studies but also serves as a peda-
gogical platform for students. By engaging with LFP data management,
exploratory analysis, and the application of advanced outlier detection
techniques, students gain practical insights and contribute to the formu-
lation of guidelines for researchers grappling with outlier challenges in
neuroscience. Keywords: Local Field Potentials (LFPs),Unsupervised
Outlier Detection,Cumulative Outlier Scores (COS), Electrophysiological
Data Analysis
Leave a Reply