Security threats in the Smart City Systems are becoming a challenge. These Smart City Systems, generating Big Data, are a revolutionizing application of the Internet of Things(IoT). Data Stream Mining, which is an efficient way of handling Big Data, is now of great concern. The acquired information is computationally expensive to process in terms of efficiency and runtime. Detection of suspicious activities on decentralized servers, generating and computing massive data streams requires time. Moreover, several stakeholders should be engaged to train the heterogenous malware data streams in the level of service application. Small experiments can be performed on the functionality of Batch ML on IoT datasets with available heap size resources. Among these candidate datasets, a little contribution has been already represented on the Mirai Attack. This research aims at the study of Data Stream Mining algorithms. Owing to the accuracy and interferences of the measurement, these algorithms are able to handle the non-hierarchical and unbalanced datasets similar to the Mirai Attacks. No single method can solely improve these critical standpoints. Thus, an Ensemble technique should be implemented. According to our study, a pool of meta or selective classifiers that interact based on the temporal Data Mining swiftly can outperform others. The maintainability and security concerns of such applications can be best fulfilled in meta-heuristics with the one-time scanning network approach for the recognition of the most frequent attacking pattern with the on-the-fly scheme. These are implemented in Create, Read, Update and Delete (CRUD) operations of the Big Data Systems.