Abstract: In recent years, machine learning algorithms have been extensively and successfully applied across multiple domains within nuclear physics, particularly demonstrating remarkable effectiveness in conducting trend analysis for the integral validation of nuclear data. The current domestic methodology for sensitivity-based trend analysis predominantly relies on manual procedures where specialists first identify a reference experimental sequence corresponding to a sensitive nuclide-reaction channel combination, then proceed to screen various experimental characteristic parameters including energy spectrum indices through expert judgment and empirical knowledge. This traditional approach aims to discover potential correlations between observed deviations in k_eff calculations and the sensitivities associated with particular nuclide-reaction channels. However, this conventional methodology exhibits considerable limitations in both efficiency and comprehensiveness, as the manual screening process proves inherently inefficient and fundamentally incapable of exhaustively exploring the vast combinatorial space of all possible correlations among diverse experimental features and nuclear data sensitivities. This significant challenge in efficiently performing thorough sensitivity-based trend analysis for criticality benchmark testing has consequently emerged as the primary obstacle confronting large-scale criticality benchmark validation initiatives. To resolve this pressing issue, our research implements sophisticated association rule analysis utilizing the well-established Apriori algorithm to systematically identify and extract frequent itemsets representing specific nuclide-reaction channel sensitivity combinations from the comprehensively organized benchmark experimental feature information database ENDITS-EXP. These systematically identified frequent itemsets subsequently underwent rigorous statistical verification procedures to determine whether significant linear regression trends exist between the calculated k_eff deviations and the corresponding nuclide-channel sensitivity coefficients, with detailed trend analysis graphs generated to visually represent these identified relationships. Through this innovative application of advanced data mining technology, an automated framework for conducting sensitivity-based criticality benchmarking trend analysis was successfully established and demonstrated, achieving substantial improvements in analytical efficiency while simultaneously refining the investigative granularity to enable more detailed and systematic examination of nuclear data performance across varied experimental conditions. This transformation from manual expert-dependent screening processes toward automated pattern discovery and correlation identification represents a crucial advancement in nuclear data validation methodology, potentially leading to more reliable nuclear data libraries and enhanced safety assessments in practical reactor physics applications. The systematic implementation of association rule mining for exploring complex sensitivity trends not only optimizes the overall analytical workflow but also reveals previously unrecognized correlations between nuclear data uncertainties and computational biases, thereby generating valuable insights for future nuclear data evaluation and refinement efforts while establishing a robust foundation for managing increasingly large and complex benchmark datasets.