{
  "schema_version": 1,
  "task_type": "regression",
  "target": "o3_average_values",
  "target_unit": "ppb",
  "features": [
    "climatic_zone",
    "alt",
    "water_25km",
    "grasslands_25km",
    "croplands_25km",
    "urban_and_built-up_25km",
    "no2_column",
    "population_density",
    "nightlight_1km"
  ],
  "id_column": "id",
  "split": {
    "column": "dataset",
    "train": "train",
    "val": "val",
    "test": "test"
  },
  "categorical_features": [
    "climatic_zone"
  ],
  "missing": {
    "policy": "train_median_mode",
    "sentinels": [
      -999
    ]
  },
  "models": [
    {
      "id": "tree",
      "learner": "regr.rpart",
      "parameters": {
        "cp": 0.01,
        "maxdepth": 4
      }
    },
    {
      "id": "forest",
      "learner": "regr.ranger",
      "parameters": {
        "num.trees": 32,
        "max.depth": 8,
        "min.node.size": 10
      }
    }
  ],
  "selection_metric": "rmse",
  "seed": 20261027,
  "row_limits": {
    "train": 400,
    "val": 120,
    "test": 120
  },
  "explanation": {
    "features": [
      "alt",
      "population_density",
      "no2_column"
    ],
    "background_rows": 40,
    "validation_rows": 60,
    "permutation_repeats": 3,
    "pdp_points": 7
  },
  "requirements": "Compare bounded tree models for station-level ozone regression; preserve the published split.",
  "unseen_categories": "train_mode",
  "analysis": "prediction",
  "feature_selection": {
    "mode": "manual"
  },
  "target_definition": {
    "sample_unit": "monitoring station",
    "averaging_period": "2010-2014",
    "prediction_time": "historical station-level metric"
  }
}