In dynamic Internet of Things (IoT) environments, traditional anomaly detection surveys often treat all anomalies as a unified concept, overlooking the distinct characteristics posed by specific anomaly types. This paper presents a structured survey and comparative analysis of anomaly detection models, organized by type of anomalies such as drift, novelty, bias, noise, constant-value, and stuck-at-zero anomalies. Each anomaly type is formally defined along with its theoretical foundation, followed by a systematic review and analysis of how model effectiveness varies across these types to identify techniques best suited for each. Our findings emphasize the need for interpretable, adaptive, and type-aware anomaly detection systems and outline open challenges in unified benchmarking, cross-type detectors, and ontology development for anomaly classification. A novel contribution of this work is a broader mapping framework that illustrates how various models differ in their ability to detect specific anomaly types across different IoT domains, offering insight into the generality and specialization of current detection approaches.