Abstract: Machine learning (ML) has emerged as a transformative approach for advancing fiber laser technology, offering powerful solutions to overcome the limitations of traditional design, optimization, and control methods. This review systematically examines the integration of ML across the entire fiber laser ecosystem. It begins by categorizing fundamental ML paradigms, with a discussion of their respective applicability. The subsequent sections detail recent research progress in key areas including intelligent device design, which encompasses ML-assisted optimization of doped fibers, photonic crystal fibers, anti-resonant fibers, polarization-maintaining fibers, fiber gratings, and mode-selective couplers; laser simulation and prediction, focusing on models for power, temporal dynamics, and spectral evolution; intelligent control of laser output, covering adaptive mode-locking, coherent beam combining, and spatiotemporal pulse shaping; and laser characterization, highlighting ML-enhanced techniques for temporal pulse measurement, mode decomposition, and beam quality evaluation. The review further addresses prevailing challenges such as data dependency, model generalizability, interpretability, and computational efficiency, while outlining future directions toward developing more robust, efficient, and physically interpretable ML-driven fiber laser systems.