Media budget allocation remains a persistent challenge in the advertising industry. Inefficient spending and biased planning decisions often reduce campaign effectiveness. Advertisers struggle to balance investments across television, radio, press, and digital platforms while managing diminishing returns. This study proposes a data-driven media mix determination model that integrates supervised machine learning with genetic algorithm-based optimization. The objective is to maximize audience reach while maintaining cost efficiency. Unlike traditional media mix models that rely on aggregated medium-level performance, this study adopts an audience exposure-based modelling approach. Facebook and YouTube are used as digital media platforms in this study. Television and digital models are trained using exposure-based reach measures, such as 1 plus and 2 plus reach. Machine learning models, including decision trees, random forests, XGBoost, and LightGBM, are evaluated to capture complex and nonlinear relationships between spend and exposure-based reach. Smoothed reach response curves are used to identify efficiency levels and saturation points for each medium. A genetic algorithm is then applied to derive the optimal budget allocation across media under efficiency, reach, and cost constraints. The model is trained using real advertising data from the Sri Lankan market, ensuring practical relevance and applicability. Although the analysis is based on a country-specific dataset, the model is transferable to markets of similar scale. This study contributes to the literature by introducing an exposure-driven media mix modelling approach that improves media budget planning accuracy and supports more effective advertising decision-making.
Dulara et al. (Thu,) studied this question.