OPMAX learning agent optimizing performance max campaigns using random forest algorithm

In this paper a new OPMAX Learning Agent has been proposed, which was trained to find the most appropriate Target Return on Ad Spend (Target ROAS) for all Performance Max campaigns within a Google Ads account, based on specific requirements of a particular user.For this purpose, a new structure of the agent is proposed, and the Random Forest Algorithm is used for its training.The machine learning algorithms used in Google Ads relay on a lots of signals during in their analyzes and observation, but there are additional requirements from the advertisers that are hard to be reached by only adjusting the existing settings of the ad campaigns.The campaigns need to be updated with different values in order to meet these goals.It is very difficult and time consuming for a person to predict what should be the exact change in the Target ROAS setting, in order the campaign to meet user's goal.The OPMAX Learning Agent will be used to predict the most appropriate value of the Target ROAS, based on the goals it has, and then it will update them in the Google Ads platform.It will work in parallel with smart bidding strategies of Google Ads, and it will make changes to the campaigns if the agent decide that an update is needed.Applying more appropriate Target ROAS improves the overall account performance and aims to achieve the desired return on ad spend.