Enhancing Retail Strategies Through Anomaly Detection in Association Rule Mining

Association rule mining (ARM) is a fundamental technique for uncovering meaningful patterns and relationships within retail datasets, providing valuable insights for decision-making processes in the retail industry. Traditional association rule mining(TARM) methods sometimes fail to handle inconsistencies and contradictions for Boolean logic concepts. To overcome the scenario, paraconsistent Annotated(PAL) logic offers a solution by embracing contradictions and providing a framework for reasoning with inconsistent information. Our approach integrates PAL with Association rule mining to form a para-association rule. Based on Boolean logic, TARM may struggle to handle such situations effectively. We propose a comprehensive framework named Para-Association rule mining (PARM), which provides decision-making by offering two distinct criteria: decided and undecided. To resolve undecided rules, an anomaly detection technique based on Isolation Forest is employed, ensuring that all rules are ultimately categorized into either "accept" or "reject" based on refined decision criteria. After solving the anomaly within the undecided criteria, all decisions are categorized into two criteria, such as accept and reject to enhanced decision-making. Our proposed approach is validated with our experimental results, where we have considered 549904 input retail transactions for generating retail decisions. Initially, we successfully generated 5000 association rules from this input, which are categorized as accept and reject 3387 and 1613, respectively. The comparative accuracy of the proposed approach archive an accuracy of 90% which is 15.38% mprovement over the traditional approach. Our approach gives an 88% inconsistency management rate, which is a 36.67% improvement over the traditional approach. To validate the robustness of PARM, a noise test was conducted, demonstrating that the model effectively maintains performance under noisy conditions. The Z-test results confirm that the robustness improvement is statistically significant. In conclusion, PARM represents a significant advancement in data mining and decision-making techniques by offering a novel framework for handling inconsistencies and uncertainty in retail datasets.