Video ads conversion rate performance modeled as classification problem

Abstract

Digital advertising has become ubiquitous thanks to the broader usage of mobile devices. Hence, advertisers require robust tools to optimize campaigns by distributing videos with high conversion rates. This work approaches the video conversion rate performance as a classification problem by studying the application of artificial neural network architectures and transfer learning for video classification. The research explores recent academic methodologies. Furthermore, the investigation undertakes multiple experiments with extended image classification networks, recurrent neural networks, and novel 3-dimension convolutional neural networks. In like manner, analyzes audio cues to find a possible correlation to video performance. Finally, the study compares and analyzes relevant evaluation metrics, like accuracy, sensitivity, and specificity, from a business perspective analyzing in-depth operational and financial risks. The proposed framework achieves significant results, over 69% of accuracy, demonstrating that it is feasible to apply deep neural networks to classify the video performance; it also opens alternatives to improve the models' performance and builds the foundation for more research in an exciting topic.