Predictive Analytics in TV Marketing For the Mass Segment

Correct TV watching rating forecasting is an important component of creating a successful advertising campaign. An advertising campaign in the television industry consists in planning the placement of advertising blocks to achieve specific goals of the advertiser's coordinated marketing strategy. Advertising blocks should be displayed at the time most appropriate to the preferences of the target audience. Today using empirical knowledge to make a forecast is less efficient than using machine learning techniques. This article describes using two machine learning models to predict TV rating and compare their results with prediction based on empirical knowledge. Recurrent neural network with Long short-term memory units and additive forecasting model were used. Article include data preprocessing steps, evaluation metric and brief description of used methods. All tests were performed using real world data from customer database. In conclusion, the directions of future work on improving the results of forecasting are given, such as improving preprocessing and enriching data.