<![CDATA[Athletes’ emotional, cognitive, and motivational states can change rapidly, yet conventional assessment often relies on intermittent self-report. This systematic review synthesized evidence on (1) psychological states targeted by machine learning (ML) and multimodal artificial intelligence (AI), (2) data modalities and model architectures used for prediction, and (3) methodological maturity of the evidence base. Following PRISMA 2020, Scopus was searched in TITLE-ABS-KEY fields for studies published from 2020 to 2025. Of 113 records identified, 4 duplicates were removed, 109 titles/abstracts were screened, 25 full texts were assessed, and 10 studies met the eligibility criteria. Screening used predefined criteria; no independent duplicate-screening record was retained, so a retrospective inter-rater coefficient could not be calculated. Evidence was synthesized narratively through thematic coding and descriptive bibliometric analysis, with study reporting quality appraised using a four-domain FICO rubric (0–2 per domain; retention threshold ≥4/8). Four themes emerged: multimodal emotion recognition, deep hybrid architectures, continuous/personalized affective-state prediction, and limited ecological validation. Multimodal fusion generally outperformed unimodal approaches, but small samples, heterogeneous labels, limited external validation, and weak interpretability constrain confidence in reported accuracy. Future work should prioritize participant-level validation, standardized psychological constructs, explainable models, and prospective field testing.]]>
Copyrights © 2025
