Data-driven maintenance decision-making can support early failure detection, reduce equipment downtime, and improve resource allocation. However, the effective application of machine learning depends not only on algorithm selection but also on technical, data-related, human, and organizational factors. This study aimed to identify and explain the factors affecting the application of machine learning in maintenance decision-making. This applied qualitative study employed a systematic literature review and thematic analysis. Studies published from 2020 through 2025 were searched in selected databases. Of 63 initially identified records, 28 articles were included in the final analysis after duplicate removal, title and abstract screening, full-text assessment, and methodological quality evaluation. Relevant statements were extracted, coded, and organized into themes and subthemes based on conceptual similarity. The findings identified 13 subthemes within eight main themes: equipment technical condition, cost and downtime, technical support, organizational structure, technology and data, resource management, predictive analytics, and systems integration. The approaches reported in the selected studies were also classified into nine algorithmic groups. The results indicate that deploying machine learning in maintenance is a socio-technical process whose effectiveness depends on data quality, sensor infrastructure, employee skills, resource management, and integration of model outputs with organizational systems. The reporting frequency of algorithms does not demonstrate performance superiority. Model selection should therefore be based on the problem, data characteristics, and operational conditions.