Interval-valued decision information system serves as a pivotal framework in the realm of decision-making processes, offering a robust structure for handling uncertainty and imprecision by encapsulating data within interval values, thereby enhancing the reliability and accuracy of decision outcomes in complex and uncertain environments. However, due to its more complex data format compared to general information systems, classical rough set models cannot be directly applied, while k-nearest neighborhood based rough set models are extremely sensitive to parameter k due to noisy data. Therefore, this study proposes a novel weighted k-nearest neighborhood rough set approach for interval-valued decision information system. We construct a weight of object x with respect to similarity of sample x to its similar and dissimilar samples in k-nearest neighborhood based on sample distance and their standard deviation. Then, the rough set model, approximate precision and approximate quality in interval-valued decision information system are addressed. Finally, a series of experiments are conducted on 8 UCI datasets, the classification accuracy, approximate precision, and quality between KNN and WKNN, the sensitivity of the model to parameters are compared. The results indicate that the proposed method is an effective candidate way to take knowledge discovery for interval value dataset.