Semiconductor devices must be characterised and measured to ensure that they perform in accordance with predefined specifications. The manual process of data-driven segregation of devices to detect anomalies is laborious and time-consuming. Therefore, there is an unmet need to automate data classification tasks in order to reduce the extensive manual review process. In order to address the issue of classifying MOSFET device characteristics, this paper explores the real-valued negative selection algorithm (RNSA) and the conservative self-pattern recognition algorithm (CSPRA), and proposes a CSPRA-SHAP classifier based on Shapley Addition interpretation (SHAP The results demonstrate that the CSPRA-SHAP classifier achieves significantly higher recall rates and accuracy in detecting and classifying anomalies than the traditional model.