Adaptive Budget Optimization for Multichannel Advertising Using Combinatorial Bandits

Optimizing budget allocation is vital for digital advertising, yet practical algorithms remain scarce due to limited public datasets and realistic simulation environments. While multi-armed bandit (MAB) algorithms are well-studied, they struggle in non-stationary settings requiring rapid adaptation. This paper introduces three key contributions: (1) a simulation environment that emulates multichannel advertising campaigns using logged real-world data; (2) an enhanced combinatorial bandit strategy with efficient exploration, and change-point detection to adapt dynamically to market shifts; and (3) Empirical validation showing superior performance over baselines in reward and regret metrics across real-world campaigns.