Smart cities increasingly rely on vast networks of IoT sensors to monitor their infrastructure, environment and public services in real time. But once those sensors are put to use in the real world, a lot can go wrong. Hardware malfunctions. Connections break. Out of nowhere, bizarre data can sometimes manifest. All that chaos can be a major disturbance on the reliability of the system. To solve that, we propose a novel distributed self-fault detection framework for smart city sensor networks. It uses hybrid K-Nearest Neighbors (KNN) approach and the thought is quite simple each sensor inspect itself i.e. pulling features from its own data and after that cross verifies any suspicious findings with neighbors. And when combined distance based KNN analysis and adaptive thresholding, the system improves problem detection capability continually under scenarios of dynamic environments or network conditions. More importantly, our distributed solution saves on the amount of information that sensors need to send, scales much better than old-school centralized solutions which push everything into one endpoint, and is far more energy-efficient. We stress tested it with a slew of simulations think permanent node failures, intermittent hiccups of differing natures, strange data bursts and the results are strong. The novel approach detects more faults, leads to fewer false alarms and saves more energy than both traditional central approaches as well as the standard configurations of KNN. All in all, it has a great trade-off between accuracy, scalability and energy use making it well suited to large smart city IoT networks.