Everyday things have made their way across the Internet to enable remote control and monitoring, covering the umbrella of Ubiquitous Computing, and thus, providing billion of smart toys for malicious actors to play with. Such networks of IoT devices facilitated with one-to-one communication are established with inherent vulnerabilities, thus, leading to zero-day attacks. These are the hitherto undiscovered threats alien to security teams with no patches and time available to trade with. The zero-day attacks become difficult to detect with conventional mechanisms (signature-based detection) as they require prior knowledge of such attacks. The learning-based techniques proved to be a promising solution to prevent the obsession to cope with these constraints motivated by these facts, this work centers around the zero-day attacks and the efficacy of learning-based techniques to detect them.