Athletes are subjected to persistent physiological and psychological stresses; however, it is a challenge to monitor accurately, in real, and at the larger scale their affective and mental conditions. Traditional self-report and clinical instruments are intrusive, retrospective, and prone to reporting bias which have triggered the use of artificial intelligence (AI) and machine learning (ML) in sport psychology. The present systematic review integrates the findings of AI- and ML-based detection, monitoring, and assessment of athletes' psychological and emotional states. APRISMA2020 process was followed to identify six hundred and twenty-five records from a pool of articles obtained via the Scopus database, which were then filtered based on the predefined eligibility criteria. In the end, after the elimination of articles via the title/abstract screening or the full-text article evaluation, 21 empirical studies from the years 2021 to 2025 were the ones to have been identified and included. For the purposes of quality and validity, the articles were judged by Mixed Methods Appraisal Tool and the results were pooled thematically. The research results revealed that when combining various physiological signals with classical ML or deep learning, recognition of emotions, stress, mental fatigue, and mental-health risk can be reached with accuracy above 85%. A major direction for future work and translation of these technologies is real-time wearable-enabled feedback. The critical issues identified were external validation, ecological realism, ethical governance, standardised datasets, longitudinal designs, and transparent and privacy-preserving deployment.