Machine learning (ML) methods have garnered increasing attention in the data science community due to their well-documented predictive performance and computational efficiency in the modern era. Many ML techniques also have robust statistical foundations and well-established operating characteristics. Despite their popularity, there remains a noticeable dearth of literature addressing the formal planning of sample sizes for predictive analytics studies employing ML methods, particularly in scenarios where access to extensive data sources is limited. This challenge becomes even more pronounced when designing prospective studies where limited resources can only support moderate sample sizes. In this comprehensive review article, we aim to examine existing methods for addressing this issue and explore potential future directions for resolving sample size dilemmas in ML-based predictive analytics.