Pleasantness in Eventfulness? Bayesian circumplex assessment for ISO soundscape prediction models

Many studies on predictability of soundscape perception apply the ISO/TS 12913-3 circumplex to derive Pleasantness and Eventfulness scores as targets in statistical models. The standard assumes a fixed geometry; its validation within prediction pipelines, however, remains uncommon. Therefore, a Bayesian framework was applied to three datasets under a shared set of acoustic and psychoacoustic predictors to assess ISO circumplex compliance. Bayesian mixed-effects ordered-logistic models were fitted to predict ratings of each perceived affective quality (PAQ) item and evaluated using participant-level cross-validation. Results showed a consistent hierarchy of learnability across datasets. Furthermore, circumplex angles were inferred with uncertainty, and axis-resolved dominance of predictor loadings on PAQ items was quantified, allowing direct comparison to ISO anchors and evaluation of cross-axis coupling in the predictor-informed space. Systematic deviations of item angles together with dominance patterns indicated increased susceptibility of the Eventfulness coordinate to mixing with Pleasantness when fixed ISO angles are used, given the considered predictors. The findings indicate that Eventfulness scores computed with fixed ISO angles can be predicted well even when the dimension is not expressed as an axis-pure acoustic construct. More generally, the proposed framework can be used to evaluate circular models of affect and supports cautious interpretation of derived dimension targets.