Key points are not available for this paper at this time.
In neuroscience and psychology research, the Pearson correlation coefficient is widely used for feature selection and model performance evaluation, particularly in studies examining relationships between brain activity and psychological behavior indices. However, when predicting psychological processes using connectome models, the Pearson correlation has three main limitations: (1) it struggles to capture the complexity of brain network connections; (2) it inadequately reflects model errors, especially in the presence of systematic biases or nonlinear error; and (3) it lacks comparability across datasets, with high sensitivity to data variability and outliers, potentially distorting model evaluation results. To better assess model performance, it is crucial to combine multiple evaluation metrics, such as mean absolute error (MAE) and root mean square error (MSE), which capture different aspects of model quality. Additionally, baseline comparisons, such as using the mean value or a simple linear regression (LR) model, provide an essential reference for evaluating the added value of more complex models. This approach offers a more robust and comprehensive analysis of functional connectomes and psychological processes.
Fu et al. (Tue,) studied this question.