Recruitment process serves as a critical biological foundation for sustainable resource maintenance. Understanding and accurately predicting the variability in recruitment has become a core challenge in fisheries management. Previous methods for predicting the recruitment of South Pacific albacore (Thunnus alalunga) have challenges in handling multicollinearity among environmental variables and identifying key drivers, often leading to model overfit and reduced predictive accuracy. To address these issues, this study applied the LASSO regression algorithm to optimize variable selection and improve prediction accuracy. Models were developed based on observational data from 2000 to 2017, and then coupled with CMIP6 multi-model climate projections, to predict recruitment trends from 2018 to 2100.The results indicated that LASSO effectively eliminated variable redundancy through shrinkage estimation, enhancing model prediction accuracy. The optimal model explained 45.9% of variance, with sea surface temperature (SST) and mixed layer depth (MLD) identified as critical predictors. Projections revealed that under high-emission scenarios (SSP5-8.5, SSP3-7.0), the population recruitment by the 2070s would approach the ecological threshold lower limit (near zero), significantly elevating collapse risks. In contrast, under the low-carbon pathway (SSP1-2.6), recruitment exhibited persistent decline trends. This study provided with an effective methodological framework for fisheries stock prediction and variable selection, while establishing a climate-recruitment coupled prediction model to provide quantitative decision-making support for formulating adaptive management strategies and mitigating population collapse risks.