Identifying Pesticide Co-Occurrence Patterns in European Soils: A Regional Analysis using Association Rule Mining (ARM)
Bekir Fatih Kahraman
article (Uluslararası, Hakemli) | Yazar(lar): Bekir Fatih Kahraman | Yıl: 2026 | Dergi/Kurum: WATER AIR AND SOIL POLLUTION | İndeks: WOS.SCI | DOI: 10.1007/s11270-026-09855-w | Anahtar Kelimeler: FP-growth, Soil monitoring, Pesticide mixtures, Machine learning, Big data analytics — Özet: This study presents an original
This study presents an original methodological approach that synthesizes Association Rule Mining (ARM) and statistical methods to decipher complex pesticide associations in European soils. The developed workflow first integrates Wilson score confidence intervals as a conservative screening heuristic for low-frequency noise elimination and subsequently employs the Frequent Pattern Growth (FP-Growth) algorithm to ensure the extraction of structurally robust co-occurrence patterns. Analyses conducted using the LUCAS 2018 soil monitoring dataset identified stable chemical bundles valid at the regional (NUTS2) scale by removing noise from high-dimensional and sparse data. The findings revealed pesticide associations in European soils characterized by a systematic chemical signature. The key patterns that emerged were aged contamination, reflecting the environmental persistence of banned substances; the accumulation of triazole fungicides resulting from agricultural use regimes; and the temporal bridge effect caused by the accumulation of seasonal applications in the soil matrix. Blind validation of the raw dataset confirmed the strong internal validity of the developed workflow and its ability to transform complex environmental monitoring data into meaningful risk management strategies. This work contributes to the scientific transition from insufficient single-substance assessments toward more realistic multi-residue scenarios, offering a methodological foundation to characterize the systematic chemical landscape of agricultural soils. By mathematically defining regional co-occurrence patterns, the proposed methodology provides a quantitative basis to enhance operational efficiency. It enables a rational narrowing of target substance lists from the broad monitored spectrum to the identified core stable groups, allowing for the prioritized allocation of limited resources to high-risk areas.