Predicting the impact of advertising: a neural network approach

This paper studies if neural networks using the temporal structure of the domain can raise the accuracy when predicting the outcome of investments in advertising (both on monthly and yearly basis), compared to the methods used today. The focus has been to investigate if future publicity can be predicted from historical outcome and planned future media investments. The domain is the car industry. This paper contains a case study where ANNs utilize time series effects for accurate prediction (the effect of advertising has a temporal signature). It is a comparative study between different network architectures that conclusively show that sequential and recurrent approaches exploit the time series dependencies and yield a performance, which supersede approaches traditionally used