Beyond beneficial or ineffective: three-way decision modeling of image-level affective responses to restorative environment design

Restorative environments cannot be defined by positive and negative emotion alone. In Attention Restoration Theory, restorative quality depends on being away, fascination, extent/coherence, and compatibility; in Stress Recovery Theory and biophilic design, physiological, safety-related, behavioral, and user-state evidence also matters. Yet, design practice often translates multidimensional evidence into binary beneficial/non-beneficial judgments. Real environments may improve mood while increasing perceived unsafety, reduce arousal while producing boredom, or provide naturalness while remaining socially uncomfortable. To address this problem, this paper proposes the Three-Way Adaptive Restorative Exposure Model (3W-AREM), integrating three-way decision theory, granular computing, and cognitive-aesthetic valuation. The full model defines successful restorative exposure as benefit across affective, attentional, perceptual, physiological, safety, behavioral, and user-state evidence without unacceptable discomfort, overload, unsafety, or avoidance. Cost-sensitive thresholds classify exposure conditions into positive, boundary, and negative regions supporting acceptance, adaptive regulation, and rejection or structural redesign. Granular computing decomposes exposure into medium, environmental type, spatial configuration, sensory composition, parameter intensity, and user-state adaptation. A secondary image-level affective demonstration used an open dataset from a peer-reviewed study on small-scale green infrastructure and affective well-being in urban sites. Because the dataset contains affective image ratings but not perceived restorativeness, ART qualities, attention tasks, physiology, safety ratings, mobility, or embodied exposure, the empirical section operationalizes only one evidence layer of 3W-AREM. Image-level positive and negative affect were used to construct fuzzy membership in an affective-restoration class. Under the main rule, 99 urban-site image conditions were classified into three positive, 87 boundary, and nine negative cases. Green-intervention images showed higher membership than control/placebo images, but most remained in the boundary region. These findings illustrate, rather than validate, that many urban green interventions are mixed affective exposure conditions requiring adaptive diagnosis, optimization, and re-testing.