Soft magnetic alloys play a critical role in power conversion, magnetic sensing, magnetic storage and electric actuating, which are fundamental components of modern technological innovation. Therefore, the rational design of soft magnetic alloys holds substantial scientific and commercial value. With excellent comprehensive performance, emerging compositionally complex alloys (CCAs) with high chemical complexity have garnered significant interest. The huge composition search space of CCAs provides both challenges and opportunities for discovering new high-performance magnetic materials. The traditional alloy design method relying on scientific intuition and a trial-and-error strategy could be inefficient and costly for magnetic CCAs. Accordingly, with great capacities for nonlinear and adaptive information processing, machine learning (ML) has shown great potential in magnetic CCA studies. This paper reviews magnetic properties of CCAs, examines the various inspiring applications of ML methods in magnetic CCAs, and discusses the future directions for unleashing the full potential of ML methods for applications in magnetic CCAs' studies.