Machine Learning-Based Classification of the Traffic of Digital Marketing Campaigns

Digital marketing i.e., any type of advertising proposed on electronic devices over the Internet, has been the prevalent advertising methodology in the last decades. As such, measuring its effectiveness and distinguishing genuine conversions generated by real users from fraudulent traffic (for example generated by bots) is of utmost importance. In this paper, such issue is tackled by using Machine Learning and, specifically, comparing a Shallow Neural Network and a Random Forest, to automatically classify traffic data of digital marketing campaigns into “good” and “bad” traffic. The two proposed models are evaluated on real traffic data, manually annotated. The results demonstrate the feasibility of the proposed classification task, with the Shallow Neural Network achieving a 98% precision and recall, without exhibiting overfitting, over the 32,000 samples of the tested dataset.

I. Introduction

Digital marketing has become one of the most used and prevalent methods of advertising in the last couple of decades with the diffusion of the Internet and especially of social media, surpassing other methods such as television, radio and magazines [1]. Specifically, in 2023, 63% of marketers say their biggest content challenge is driving traffic and generating leads, whereas traditional marketing generates 50% fewer interactions with customers than digital marketing [2]. The process of marketing has fundamentally changed over the years, going from general advertisements that were aimed at reaching a wide audience to personalized and targeted ads and messages for every single user, a process known as “People-based marketing” [3]. For example, in 2016, almost 63% of advertisers said that they had seen improved click-through rates with addressable media and 60% had experienced higher conversion rates [4].