Various social engineering (SE) tactics are used to trick users into divulging personal or sensitive information, purchasing fake products, and downloading software. These types of socially engineered malicious ads are published on the low-tier ad networks, where chance of security check is less. In this research study, we propose a custom-built deep learning CNN-based framework for the Identification of SE Attack Campaigns delivered via Malicious Advertisements (SEACMA). In our framework, a concept of federated learning is implemented on top of transfer learning and is trained on data collected through research on discovering and tracking attack campaigns to identify multiple attacks campaigns as the Malicious Ads. SE attacks have strong visual components which help in the generalization of the network and achieving our main goal of detecting SE attack web pages by classifying page screenshots. The results show that the framework can identify the malicious ads with impressive detection scores.