TV Network Audience Share Prediction
Mediapro, the leader in television production in Spain, faces the challenge of predicting the success of its own programs and those of its competitors. This analysis is crucial since predicting the audience level of a channel determines advertising costs at different time slots. In its quest to lead technological innovation, Mediapro collaborated with several firms, including WhiteBox, for improving its audience prediction methodology.
At WhiteBox, we chose to move beyond conventional methods of time series analysis, such as ARIMA and SARIMA, in favor of more innovative and less computationally demanding solutions, adapting to existing infrastructure limitations. We opted for the use of Darts, a cutting-edge, open-source library for time series analysis, and we opted for a robust algorithm based on decision trees (gradient boosting) for its efficiency, simplicity, and ease of scaling. Before creating our model, we carried out a meticulous data cleaning process in collaboration with Mediapro, ensuring a solid base for model training.
The implementation of our model has provided Mediapro with a novel perspective to improve its audience prediction system, standing out for its power and ease of use both during training and in practical production use.