Neon flying squid Ommastrephes bartramii is a cephalopod species of economically importance distributed in the North Pacific Ocean. Because of short lifespan, their abundances mainly depend on the recruitments and the marine environment during their early life stages will directly affect the recruitments. Using fishery data collected by Chinese squid-jigging fleets in the Northwest Pacific Ocean from 2004 to 2015 and the sea surface temperature (SST) of spawning and feeding grounds, which were divided into different numbers of subareas, correlation analysis and random forest model were used to screen out the subareas that CPUE has significant relationships with Ps (the proportion of favorable-SST in spawning grounds) of spawning grounds and Pf (the proportion of favorable-SST in feeding grounds) of feeding grounds during spawning and feeding periods. The Ps and Pf were used as input variables of the neural network model to forecast recruitments based on spawning ground and feeding ground, respectively, and the advantages and disadvantages of the model and forecast accuracy were analyzed. The results show that the schemes of dividing spawning ground into 5°×5° and feeding field into 2.5°(longitude)×4°(latitude) were optimal. The ranges of subareas selected by random forests are largely consistent with those selected by correlation analysis, both random forests and correlation analysis can identify potential subareas associated with CPUE, and both have forecasting accuracy of >90%. However, the subareas selected by the random forest are better and the forecast accuracy is higher than those selected by correlation analysis. In addition, the model based on the spawning ground is more accurate and stable than that based on feeding ground.