Abstract: Aluminum alloys feature light weight, high specific strength, favorable corrosion resistance and outstanding fatigue properties, and possess wide application prospects in marine equipment. Accurate prediction of fatigue crack growth rate (FCGR) at various crack propagation stages is crucial to evaluate the service life and structural integrity of key load-bearing components of marine equipment. Conventional physics-based theoretical models and empirical equations generally fail to accurately capture the complicated nonlinear relationship between fatigue crack growth rate (da/dN) and stress intensity factor range (ΔK). Furthermore, the influences of stress ratio (R) and temperature (t) on the FCGR of aluminum alloys are remarkably affected by material composition and manufacturing processes, which imposes challenges to the generality of existing crack growth rate and life prediction models. In this work, fatigue crack growth rate tests were firstly performed on 7B05 aluminum alloy. Three-stage FCGR curves under different t and R were acquired, and the effects of temperature and stress ratio on crack growth rate were investigated. Published aluminum alloy FCGR data from literatures were further collected and fused with the above experimental data to establish a dataset of aluminum alloy fatigue crack growth rate. The effectiveness of Extreme Gradient Boosting (XGBoost) and Random Forest (RF) models for predicting FCGR covering all three stages was compared respectively. The results show that prediction outputs of the two machine learning models are unsatisfactory when trained merely with literature data. Nevertheless, the introduction of experimental data of the target material greatly improves prediction accuracy, which highlights the significance of such data for aluminum alloys applied in marine engineering. As the XGBoost algorithm achieves optimal prediction performance, a wider range of independent variables were incorporated and hyperparameters were optimized. The optimized model realizes high accuracy in predicting FCGR under various t and R conditions, so as to reduce dependence on extensive and expensive experimental programs. These findings not only deepen the understanding of fatigue crack propagation of this specific aluminum alloy, but also offer a more universal method to predict crack growth of other materials subjected to extreme ambient temperatures and complex stress ratios.