P layer R etention P rediction M ethod for F reemium G ames

Player retention in freemium games is crucial to the users' lifecycle and operational revenue. This article uses the fixed player queue in Brawl Stars as a case study to construct a reproducible workflow covering battle log cleaning, time window division, player-level feature engineering, and classification model evaluation. This article uses 46 characteristics measured within 24 hours after the player's first baseline observation-including battle count, active span, battle interval, win rate, trophy changes, mode participation, and hero information-to predict whether the players will be active again within the 7th to 8th day window. The latest data includes 166 players and 41,402 post-cleanup battle records. The D7 tag override check passed, with 99 players active in the D7 window and 67 inactive. After player grouping, random forest achieved the highest ROC-AUC of 0.661, accuracy of 0.706, recall of 0.900, and F1 value of 0.783. The results show that early combat behavior has some D7 retention differentiation ability, but the current sample size and a single game source limit the external validity of the conclusions.