π Classification Thresholds, Validation & Regularization
This collection focuses on decision thresholds, model evaluation, cross-validation, and regularizationβthe pillars behind the interactive experiences that balance clinical risk, assess generalization, and control model complexity.
Table of Contents
- Evaluation Metrics & Thresholding
- Cross-Validation & Resampling
- Regularization & Generalization Control
- Healthcare Case Studies
1. Evaluation Metrics & Thresholding
| Resource | Focus | Access |
|---|---|---|
| Google ML Crash Course β Thresholds and the Confusion Matrix | Illustrated guide to shifting thresholds and interpreting error trade-offs. | π¬π§ English |
| Google ML Crash Course β Accuracy, Precision, Recall | Definitions and intuition for the metrics surfaced in the Risk Manager demo. | π¬π§ English |
| Google ML Crash Course β ROC and AUC | Explains ROC analysis, tying directly to balancing false negatives and false positives. | π¬π§ English |
| scikit-learn Model Evaluation Guide | Comprehensive reference covering confusion matrices, ROC, precision-recall, and calibration. | π¬π§ English |
2. Cross-Validation & Resampling
| Resource | Why it matters | Access |
|---|---|---|
| scikit-learn Cross-Validation Overview | Describes K-Fold, Stratified, ShuffleSplit, and nested cross-validation. | π¬π§ English |
| Stone, M. (1974). Cross-Validatory Choice and Assessment of Statistical Predictions. | Classic paper formalizing cross-validation for model assessment. | https://projecteuclid.org/journals/journal-of-the-royal-statistical-society-series-b/volume-36/issue-2/Cross-Validatory-Choice-and-Assessment-of-Statistical-Predictions/10.1111/j.2517-6161.1974.tb00994.x.full |
| Arlot, S. & Celisse, A. (2010). A Survey of Cross-Validation Procedures for Model Selection. | Survey detailing when to prefer K-Fold, leave-one-out, and Monte-Carlo cross-validation. | https://arxiv.org/abs/0907.3838 |
| Kohavi, R. (1995). A Study of Cross-Validation and Bootstrap for Accuracy Estimation and Model Selection. | Empirical comparison that motivates the Honest Validator storyline. | https://dl.acm.org/doi/10.5555/1643031.1643047 |
3. Regularization & Generalization Control
| Resource | Highlight | Access |
|---|---|---|
| scikit-learn β Regularization in Logistic Regression | Explains the C hyperparameter, penalties (L1, L2, elastic net), and solver behavior. | π¬π§ English |
| Hastie, Tibshirani & Wainwright (2015). Statistical Learning with Sparsity: The Lasso and Generalizations. | Deep dive into L1/L2 penalties, shrinkage, and sparsity. | https://web.stanford.edu/~hastie/StatLearnSparsity/ |
| Ng, A. (2004). Feature Selection, L1 vs. L2 Regularization and Rotational Invariance. | Shows why L1 induces sparsity and when to prefer L2. | https://cs229.stanford.edu/notes2020spring/cs229-notes3.pdf |
| Goodfellow, Bengio & Courville (2016). Deep Learning β Chapter 7 | Conceptual overview of capacity control, regularization, and bias-variance trade-offs. | https://www.deeplearningbook.org/ |
4. Healthcare Case Studies
| Resource | Contribution | Access |
|---|---|---|
| Rajkomar et al. (2018). Scalable and Accurate Deep Learning with Electronic Health Records. | Includes logistic baselines and evaluation metrics in clinical settings. | https://www.nature.com/articles/s41746-018-0029-1 |
| Powers (2011). Evaluation: From Precision, Recall and F-Measure to ROC, Informedness & Markedness. | Provides statistical interpretation of evaluation metrics used for medical classification. | https://arxiv.org/abs/2010.16061 |
| Chicco & Jurman (2020). The Advantages of the Matthews Correlation Coefficient (MCC). | Discusses alternative metrics valuable for imbalanced clinical datasets. | https://www.nature.com/articles/s41598-020-76158-9 |
| Saito & Rehmsmeier (2015). The Precision-Recall Plot Is More Informative than the ROC Plot when Evaluating Binary Classifiers. | Supplement to ROC discussions for imbalanced medical screening. | https://journals.plos.org/plosone/article?id=10.1371/journal.pone.0118432 |
Note: All links were re-checked in October 2025. For licensed resources, rely on institutional subscriptions or open-access copies.