A Benchmark Dataset for Marine Water Quality Classification and SHAP Interpretability

DATASET PROVENANCE AND ARCHIVAL CURATION This is a comprehensive secondary marine water quality dataset containing exactly 10,000 continuous records. It was originally sourced from an open-access environmental repository on Kaggle, curated by the author 'robust_mariner'. As the original public repository has recently transitioned to a restricted status, this verified local copy is formally archived here to ensure permanent academic reproducibility, adhering strictly to FAIR (Findable, Accessible, Interoperable, and Reusable) data principles. DATA VERIFICATION AND CURATION For the purpose of this specific machine learning comparative analysis and SHAP interpretability benchmarking, the dataset was rigorously evaluated, mathematically verified as complete, and structurally balanced by the Data Curator (Yuthriya Khamis). This proactive curation directly supports ongoing Master's thesis research at Multimedia University (MMU). BIOGEOCHEMICAL DYNAMICS REFLECTED (ORIGINAL METHODOLOGY) To ensure full scientific transparency regarding the dataset's origins, the parameters capture realistic marine water quality dynamics, mathematically derived to reflect the following established biogeochemical couplings: Dissolved Oxygen (DO): Modeled with a temperature-dependent decay function capturing inverse solubility: DO_adj= DO ×exp⁡(-0.05 × (Temp – 20)) Nitrate Dynamics: Includes aerobic denitrification effects where $NO_3$ is coupled with oxygen cycles. Chlorophyll-a: Corresponds to nutrient-driven growth patterns formulated as: Chl_a = 15 + 25 x sigmoid(NO_3 - 10) + N(0, 5). CLASSIFICATION GROUND TRUTH Water instances are strictly classified as 'Safe' (1) only if they satisfy all three critical criteria: Dissolved Oxygen > 8 mg/L, Nitrate < 15 mg/L, and Chlorophyll-a < 60 μg/L. All other instances are classified as 'Unsafe' (0).