Technical Articles & Case Studies¶
In-depth investigations into tree architecture, regulatory compliance, industrial IoT, econometrics, and survival analysis. Each study is grounded in real benchmark data with reproducible code, empirical holdout metrics, and publication figures.
Featured Case Studies¶
-
Decision Trees for Credit Scoring
Regulatory compliance, path-level auditing, and probability calibration under fair lending constraints.
Dataset: Give Me Some Credit (150,000 borrowers)
-
Zero-Imputation Predictive Maintenance
Industrial sensor feeds: native missing-data routing and surrogate splits under severe asymmetric misclassification costs.
Dataset: Scania Truck APS Failure (76,000 vehicles)
-
Uplift Modeling with Causal Trees
Commercial decision optimization: honest sample-splitting (\(S^{\text{tr}}\) vs \(S^{\text{est}}\)) to identify persuadable customers without sleeping-dog harm.
Dataset: Hillstrom Randomized Trial (64,000 customers)
-
Modernizing Quinlan Classics in Rust
Re-architecting M5, Cubist, and C5.0 into high-throughput systems ML: linear leaves, dynamic rule extraction, and exact parity.
Dataset: Concrete Compressive Strength (1,030 mixtures)
-
Beyond Mean Squared Error: Distributional Trees
Actuarial & risk modeling: Tweedie compound Poisson-Gamma claims, Tobit censored MLE, and heteroscedastic regimes.
Dataset: Miami Housing & German Credit (TabArena)
-
Breaking Step Functions: Model Trees vs. GBDT
Systems ML & continuous tabular regression: why piecewise linear leaves eliminate staircase artifacts with far shallower trees than GBDT.
Dataset: Diamonds (53,940 stones)
-
Segregate First, Model Later: Two-Stage Survival Trees
Actuarial & retention analytics: decoupling ID-level tree cohort discovery from leaf-level Kaplan-Meier curves to eliminate right-censoring bias and non-proportional hazard violations.
Dataset: Bank Customer Churn (10,000 customers)
Quick Problem Navigator¶
| Business & Modeling Question | Case Study | Benchmark Dataset |
|---|---|---|
| How do I audit a classifier and choose an operating threshold? | Decision trees for credit scoring | Give Me Some Credit |
| Can a tree score industrial telemetry with missing sensor values without imputation? | Decision trees and missing sensor data | Scania APS Failure |
| Does an uplift model improve targeting without waking sleeping dogs? | Causal trees for uplift modeling | Hillstrom randomized trial |
| How do Quinlan's M5 and Cubist rule sets translate to high-speed Rust? | Cubist-style model trees | UCI concrete strength |
| What changes when I use Tweedie and Tobit distributional split criteria? | Distributional tree splits | Miami home sales |
| When do piecewise linear leaves outperform gradient boosted trees? | Regression trees with linear leaves | Diamonds |
| How do I model longitudinal customer churn without right-censoring bias? | Two-stage hierarchical survival trees | Bank Customer Churn |
Next Steps¶
- For procedural recipes and execution workflows, see the Practical Modeling Pathways.
- For focused operation scripts, browse the Example Library.
- For complete parameter definitions and methods, inspect the API Reference.