Accurately calculating catch per unit effort (CPUE) is a prerequisite for predicting fishery resource abundance and distribution. In tuna longline fisheries, the statistics of catch and effort are often influenced by spatial scale, and different sources of catch and effort data can result in biased CPUE estimates, which thereby potentially affect the forecast precision of fishing grounds. This study utilized fishery data obtained from longline vessels operating in waters near Micronesia, derived from both fishing logbooks and the VMS (vessel monitoring system), to calculate two types of CPUE for yellowfin tuna (Thunnus albacares). Using a BiLSTM (Bi-directional long-short-term memory) model, both types of CPUE were employed as inputs, alongside spatio-temporal factors and ocean remote sensing data, to develop a predictive model for yellowfin tuna fishing grounds. By SHAP (SHapley Additive exPlanations) analysis, key variables influencing the prediction of yellowfin tuna fishing grounds were identified. The spatio-temporal distribution characteristics of fishing grounds were determined based on the geographical distribution of different CPUE values. This study also evaluated the impact of different CPUE inputs on the accuracy of fishing ground prediction. The results showed that the BiLSTM model effectively predicts yellowfin tuna fishing grounds, with strong performance within the study area. Moreover, the model using CPUE calculated from VMS data demonstrated higher accuracy compared to the model built by using logbook data. Key variables influencing yellowfin tuna fishing ground predictions included chlorophyll-a concentration, dissolved oxygen concentration at 200 m, temperature at 300 m, temperature at 200 m, and longitude. Fishing grounds derived from fishing logbook were dispersed across the north and south, whereas those derived from VMS data were mainly concentrated in the southern region. Based on the vertical profile of temperature and dissolved oxygen, it was inferred that the fishing grounds identified by VMS data are more accurate. In the waters of Micronesia, yellowfin tuna fishing grounds were more densely distributed in the first and second quarters. This study demonstrates that VMS-based CPUE can be calculated for longline fisheries and effectively used for fishing ground prediction. Based on this study, it is recommended to record catch locations individually using the VMS method, when investigating tuna fishing grounds in longline fisheries at smaller spatial scales. This allows for grid-based CPUE calculations, thereby improving the accuracy of fishing ground prediction.