Multi-timepoint pattern analysis: improving classification with neural timeseries data.

Area of Science:

• Neuroscience
• Computational Neuroscience
• Data Science

Background:

• Long, naturalistic stimuli elicit complex neural responses.
• High-dimensional neural timeseries data pose challenges like overfitting and reduced predictive power due to limited sample sizes.

Purpose of the Study:

• Introduce multi-timepoint pattern analysis (MTPA) as a temporal dimension reduction technique.
• Improve prediction accuracy for models using long neural timeseries data.
• Enhance the interpretability of neural data analysis.

Main Methods:

• Developed MTPA, a temporal dimension reduction approach.
• Utilized feature selection with elastic net regression within MTPA.
• Compared MTPA against principal component analysis, windowed averaging, and no dimension reduction.

Main Results:

• MTPA consistently outperformed other methods across two experiments.
• Experiment 1 achieved up to 79.1% accuracy in predicting psychological states.
• Experiment 2 achieved up to 66.5% accuracy in predicting cognitive load and narrative context.

Conclusions:

• MTPA is a valuable tool for analyzing neural data from extended naturalistic designs.
• The method enhances prediction accuracy and provides insights into neural temporal dynamics.
• MTPA offers a way to overcome challenges associated with high-dimensional neural timeseries data.