Mining irregular patterns in environmental data using density and neighborhood analysis

Environmental data are often characterized by high variability arising from natural processes, anthropogenic activities, seasonal fluctuations, and long-term environmental change, making the reliable identification of irregular patterns a challenging analytical task. To address this challenge, this study proposes an unsupervised analytical data mining framework for detecting irregular environmental observations across heterogeneous environmental datasets. The proposed framework integrates three complementary analytical components—neighborhood deviation analysis, density sparsity characterization, and clustering-based structural consistency modeling—into a unified irregularity scoring function capable of identifying both isolated outliers and context-dependent anomalies without requiring labeled data or predefined anomaly templates. The framework was evaluated using three publicly available environmental datasets comprising air quality records obtained from CPCB and WAQI, water quality observations from WQP and CWC, and climate variability data from NOAA and IMD. Experimental results demonstrate strong detection performance, achieving detection rates of 91.8%, 88.5%, and 86.9% for air quality, water quality, and climate variability datasets, respectively, while maintaining false alarm rates below 10%. Sensitivity and ablation analyses further confirm the robustness of the framework, with F1-scores exceeding 0.85 across varying parameter configurations. Runtime analysis demonstrates near-linear computational scalability, requiring approximately 1.17–1.24 ms per observation for datasets ranging from 10,000 to 250,000 records. The present framework focuses on state anomaly detection within a multivariate environmental feature space and does not explicitly model temporal dependencies or seasonality, which remain important directions for future research.