Use of Grid Search in Hyper-Parameter Selection for Time Series Analysis: A Case Study with Ad Mediation Software
Görkem GİRAY Murat Osman ÜNALIR Şeyma TAHMAZ
AbstractThe success of an Ad Mediation Software’s decision on from which ad network to request ads and in which order depends on the ability to estimate eCPM (effective Cost Per Mille) value used to measure ad revenue. This value varies for different applications depending on different external factors. It is not possible for domain experts to make successful predictions by analyzing different sets of external factors for a large number of applications and to keep these estimates constantly up to date. Therefore, eCPM values were automatically estimated separately for each application based on different advertising spaces and different countries using time series analysis. The ARIMA model was used to estimate and the hyper-parameters of the model were optimized by using grid search method. For most of the values obtained, it was found that the values obtained with the intuition of domain experts were closer to the actual values.