Operators' lack of understanding of the plant's operation state significantly contributes to human errors in Nuclear Power Plant (NPP) control room operations. The state of an NPP at a particular time is represented by values of analog (e.g., measurements of flow properties) and switch parameters (e.g., the status of a valve). Previous studies focused on analyzing analog parameters rarely considered the switch parameters. Estimating the plant state without considering the timings of switches can be inaccurate. This paper utilizes analog parameters to infer the timing of switches. Two main challenges of establishing a reliable prediction model are 1) high dimensional analog parameters and 2) an imbalanced switch parameter dataset with few control actions. This paper uses PCA to reduce the dimensions and SMOTE to generate more samples capturing the impacts of various control actions. Then the pre-processed data was used to train variants of KNN classifiers. Testing results show that the KNN with SMOTE oversampling but without PCA best predicts switches' timing.