AI-Driven Game Optimization for Unreal Engine: Profiling-Aware Diagnosis and Auto-Refactoring

Abstract

Modern games push CPU/GPU, memory, and I/O to their limits. We propose the UE AI Optimizer, an AI layer that converts Unreal profiling traces into ranked, explainable optimization actions and safe editor auto-refactors, with an interpretable 0–100 OptiScore. This is a proposal with a preregistered evaluation plan. Under assumptions (A1–A4), our prospective targets are: recommendation Precision@5 ≥ 0.80, Top-1 Impact Hit ≥ 0.70, impact-prediction MAE ≤ 20% (ECE ≤ 0.10); scene-level p95 frame-time reductions of 15–30% for blocking loads via async/streaming, 10–25% for over-Tick logic via timers/events, and 10–20% server p95 for heavy replication via RepGraph/Iris; and +8–20 OptiScore points across three projects/hardware tiers. These are theoretical goals; we will validate with paired tests, effect sizes, and calibration analysis after implementation.