A Comparative Analysis of the Efficacy of Machine Learning Performance, Interpretability and Age-Specificity across Neuroimaging and Behavioral Data

2026-02-12T07:50:38+00:00

This study gives a comparative analysis on the performance of machine learning models, their interpretability, and age specificity on neuroimaging and behavioral data in a population in Nigeria. Using a quantitative framework, standardized cross-validation protocols were used to benchmark linear (ridge and LASSO) models, nonlinear (support vector regression) models and ensemble (random forest and gradient boosting) models. To assess how development affects predictive performance and how much the feature can be understood, an age stratification was introduced. Findings showed that the ensemble models were always of better predictive quality, especially when it comes to adult cohorts, whereas linear models are the best in terms of feature stability and interpretability. The behavioral data tended to have better predictive performance than did neuroimaging data, with lower noise-to-signal ratios and greater correspondence of features to outcomes. The age factor was also a major moderator with younger cohorts showing lower model generalization and interpretability. It was also found that there was an inverse correlation between predictive efficacy and interpretability, and this was due to the trade-offs of complex model architectures. These results highlight the relevance of context-dependent model appraisal models that combine various performance measures with interpretability and stability measures. The implications of the results to the responsible use of machine learning in neuroscience and behavioral studies in Nigeria are that it can be used in age-dependent applications.

IJO - International Journal of Applied Science (ISSN: 2992-247X )

A Comparative Analysis of the Efficacy of Machine Learning Performance, Interpretability and Age-Specificity across Neuroimaging and Behavioral Data