ML Risk Classification & Explainability

ML Data Quality Gate

action data_quality_gateon fail: abort

config (5 keys)

{
  "constructs": [
    "supply_chain_disruption_binary",
    "supplier_financial_health",
    "lead_time_variability",
    "geographic_concentration_risk",
    "network_betweenness_centrality"
  ],
  "max_missing_pct": 0.1,
  "min_observations": 200,
  "min_positive_class_pct": 0.05,
  "require_balanced_classes": false
}

Feature Engineering & Selection

action analyze

config (4 keys)

{
  "create_interaction_terms": true,
  "create_lag_features": true,
  "handle_imbalance": "smote",
  "lag_periods": [
    1,
    3,
    6
  ]
}

Random Forest — Disruption Prediction

engine random_forest←feature_engineering

config (5 keys)

{
  "cv_folds": 5,
  "n_estimators": 500,
  "outcome": "supply_chain_disruption_binary",
  "predictors": [
    "supplier_financial_health",
    "lead_time_variability",
    "geographic_concentration_risk",
    "network_betweenness_centrality",
    "geopolitical_risk_index",
    "climate_exposure_score",
    "tier_depth",
    "inventory_buffer_ratio"
  ],
  "scoring": "roc_auc"
}

expected results (1 key)

{
  "auc": {
    "min_value": 0.7
  }
}

XGBoost — Gradient Boosting Benchmark

engine gradient_boosting←feature_engineering

config (5 keys)

{
  "cv_folds": 5,
  "early_stopping_rounds": 50,
  "outcome": "supply_chain_disruption_binary",
  "predictors": [
    "supplier_financial_health",
    "lead_time_variability",
    "geographic_concentration_risk",
    "network_betweenness_centrality",
    "geopolitical_risk_index",
    "climate_exposure_score",
    "tier_depth",
    "inventory_buffer_ratio"
  ],
  "scoring": "roc_auc"
}

SHAP Feature Importance & Explanations

engine shap_explainer←xgboost_model

config (3 keys)

{
  "model_source": "xgboost_model",
  "plot_types": [
    "summary",
    "dependence",
    "force"
  ],
  "top_features": 10
}

Isolation Forest — Anomaly Detection (Unsupervised)

engine isolation_forest

config (3 keys)

{
  "contamination": 0.05,
  "features": [
    "lead_time_variability",
    "supplier_financial_health",
    "trade_flow_volume",
    "port_congestion_index"
  ],
  "n_estimators": 200
}

Causal Forest — Heterogeneous Treatment Effects

engine causal_forest←random_forest_model

config (4 keys)

{
  "covariates": [
    "geographic_concentration_risk",
    "tier_depth",
    "supplier_financial_health"
  ],
  "n_estimators": 500,
  "outcome": "supply_chain_disruption_binary",
  "treatment": "supplier_diversification"
}

Model Comparison & Recommendation

action analyze←random_forest_model, xgboost_model, anomaly_detection

config (3 keys)

{
  "compare_models": true,
  "generate_risk_scores": true,
  "recommend_production_model": true
}