Interpretable classifiers for FMRI improve prediction of purchases

Area of Science:

• Neuroscience
• Machine Learning
• Neuroimaging

Background:

• Machine learning (ML) is increasingly applied to neuroimaging, but predicting behavioral output remains challenging.
• Functional magnetic resonance imaging (fMRI) offers potential for predicting behavior due to its spatial and temporal resolution.
• fMRI data present analytical challenges including low signal-to-noise ratio, high dimensionality, and spatiotemporal correlations.

Purpose of the Study:

• To investigate the predictive power of fMRI activation in specific brain regions for purchasing decisions.
• To enhance spatiotemporal interpretability and classification accuracy in neuroimaging analyses.
• To compare the performance of sparse penalized discriminant analysis (SPDA) against other ML algorithms for predicting behavior.

Main Methods:

• Applied various ML algorithms, including SPDA, logistic regression, linear discriminant analysis, and linear support vector machines.
• Utilized previously acquired fMRI data focusing on activation in the nucleus accumbens (NAcc), medial prefrontal cortex (MPFC), and insula.
• Employed SPDA for automatic selection of correlated variables to improve model interpretability and generalization.

Main Results:

• SPDA demonstrated superior performance compared to logistic regression, linear discriminant analysis, and linear support vector machines.
• SPDA significantly improved classification accuracy in predicting purchasing decisions.
• SPDA yielded interpretable models that generalized effectively to new, unseen data.

Conclusions:

• SPDA offers a powerful approach for building interpretable neuroimaging models that predict choice behavior.
• This method allows for more precise inferences about the contribution of specific brain regions to decision-making.
• The framework provides a generalizable method for applying ML to neuroimaging for behavioral prediction.