Unified metric for heterogeneous prediction performance and model selection

Determine a most appropriate, unified metric for summarizing prediction performance across heterogeneous data types in attributed multilayer networks and ascertain how this metric should guide selection of the optimal model during cross-validation procedures when information spans different parameter spaces.

Background

The paper introduces PIHAM, a probabilistic generative model for inference in heterogeneous attributed multilayer networks. Because inputs can include multiple interaction types and attributes with different statistical supports, evaluating predictive performance with a single metric is challenging and influences cross-validation choices for model selection (e.g., choosing the number of communities K). The authors explicitly note that despite their proposed approach, the question of an appropriate unified metric remains unresolved.