Random Forest Guided Feature Weighted KNN Model for Nuclear Reactor Event Identification

Abstract

Accurate identification of developing events is crucial in achieving reactor operation safety, decision making and accident management.  Machine learning techniques are increasingly used as alternative to conventional model based methods due to their data driven nature for abnormal event detection. Among these, the K-Nearest Neighbors (KNN) algorithm remains widely used across industrial diagnostic systems, including applications in nuclear reactor monitoring, owing to its simplicity and good performance in multi class classification tasks. However, a key limitation of standard KNN is that all input features are treated with equal importance, which can limit its accuracy. To address this limitation, this study proposes a Random Forest guided feature weighted KNN classifier for reactor event identification. In the proposed approach, feature importance extracted from a Random Forest model is utilized to strengthen the ability of standard KNN model. The method is evaluated on a five class nuclear reactor event dataset and is found to outperform the base KNN and RF models. The performance of the proposed model is also compared with other KNN - RF ensemble methods and is shown to perform the best among all with superior performance metrics.