A COMPARATIVE STUDY ON EFFECT OF TIME SERIES MODELLING AND MACHINE LEARNING APPROACH TO PREDICT ADVERTISEMENT AIRING TIME INVENTORIES

Abstract

Although there are emerging streams to deliver the promotional messages to customers such as social media marketing and email marketing, television advertisements have the dominating power over them. The local and global television companies make their revenue basically by publishing these commercials or advertisements to the end user during the TV programs. Moreover, some of these local television operators borrow foreign channels and broadcast them locally. Hence, these local TV operators should have the information on program schedules of these foreign channels in advance to prepare their advertisement inventories which need to be sold to customers. However, the local TV operators usually receive the program schedule from global TV channels very close to the actual schedule date. Thus, they do not have adequate time to sell their advertisement airing time to their customers. The proposed approach of this study has addressed and achieved this problem by utilizing time series modelling and machine learning approaches such as SARIMAX, SVR, RFR, GBR and LSTM. The experimental results show that both time series and machine learning models can be used interchangeably to forecast the next seven days of advertisement airing time/ ad inventory in one hour time resolution for given TV channels with a significant level of accuracy. Furthermore, the LSTM model has shown better accuracies for five test samples with mean deviation of 89 seconds.