notes · learning path
DS & Statistics Learning Track
A structured learning path for statistics, model development, validation, and advanced risk-modelling methods. Short notes and interactive pages to see, test, and reason about otherwise abstract concepts.
Order matters: start with the modelling core, then statistical foundations, then advanced methods, and
finally model risk, decision policy, and explainability.
Reviewed: 15 July 2026Learning rule: Formulas, plots, and interpretation should be read together.Scope: Statistical explanation; thresholds are contextual, not universal.Simulator note: Synthetic examples are teaching aids, not production estimates.
Total Pages
21
A full path from core modelling and statistics to calibration, uncertainty, shrinkage, decision policy, and explainability.
Format
Notes + Tools
Some pages are concept-driven, some interactive, and many are designed as both explanation and sandbox.
Odak
Risk Modelling
The whole track is built around validation logic, PD thinking, calibration, robustness, threshold policy, and model governance.
New here? Suggested starting path
01 → 02 → 03 → 04 → 05 → 06 → 17 → 20 → 21
track 01
Start Here / Core Modelling
The fastest way to understand what a risk model is, how it is evaluated, how it is calibrated, and how it becomes a working validation workflow.
01
Regression Analysis
Linear · Logistic · OLS · Residuals · R²
The first real modelling layer: fit, coefficients, residual logic, and the difference between explanatory relationships and predictive structure.
CORE
02
Model Discrimination
ROC · AUC · Gini · KS · CAP
The ranking question: can the model separate good from bad? This page ties the main discrimination metrics together as one system.
CORE
03
Calibration, PD Scaling & Probability Alignment
Calibration Curve · Intercept · Slope · Brier · PIT / TTC
The probability layer of the model: not just who is riskier, but whether the PD figure is credibly aligned with realised frequency.
CORE
04
Model Stability & Backtesting
PSI · CSI · Drift · H-L · Backtesting
The monitoring layer: has the population shifted, has calibration drifted, is the model still behaving acceptably over time?
CORE
05
WoE, IV & Data Preparation
WoE · IV · Binning · Missing Data · Outliers
Where raw data is turned into scorecard-ready inputs. Monotonicity, information value, missing-data logic, and the structure behind sound feature preparation.
CORE
06
Model Selection
AIC · BIC · Adjusted R² · Overfitting · VIF
Why "better fit" is not enough. How complexity, parsimony, and multicollinearity shape the final model choice.
CORE
07
KS & Confusion Matrix
KS Statistic · Precision · Recall · F1 · MCC
A deeper look at threshold logic: where separation is strongest, how cutoffs change the error mix, and why decision rules are a policy choice.
CORE
track 02
Statistical Foundations
The background statistical language that makes the modelling pages deeper, cleaner, and easier to interpret.
08
Distributions
Normal · Binomial · Poisson · PDF · CDF
Intuition for uncertainty itself: shape, probability mass, tails, and what distributional assumptions really mean in practice.
FND
09
Descriptive Statistics
Mean · Median · Mode · Skewness · Kurtosis
The basic vocabulary for describing a dataset before modelling begins: centre, shape, asymmetry, and why averages can hide structure.
FND
10
Variability Measures
Variance · Std Dev · Range · IQR · CoV
Understanding spread, dispersion, and robustness. Where stability and volatility move from intuition to something measurable.
FND
11
CLT & Correlation
Central Limit Theorem · Pearson · Spearman
The first bridge from raw data to inference: why sampling stabilises, and how variables can move together without explaining each other.
FND
12
Hypothesis Testing
p-value · t-test · z-test · Type I & II Error
What statistical evidence can and cannot say. This page is critical because much of the model-validation language rests on test logic.
FND
13
Parametric Tests
ANOVA · Chi-Square · F-test · Assumptions
Move beyond one-off tests: multi-group comparison, variance logic, and the assumptions that quietly govern validity.
FND
track 03
Advanced Methods
The part of the track where time, uncertainty, rare defaults, and macro structure take centre stage.
14
Survival Analysis & Time-to-Default
Kaplan-Meier · Cox PH · Hazard · Censoring
Move from "will it default" to "when will it default". Essential for lifetime PD, censoring logic, and time-to-event thinking.
ADV
15
Time Series Fundamentals
ARIMA · Stationarity · ACF · PACF · Forecasting
The macro and monitoring side of modelling: trend, lag structure, differencing, autocorrelation, and IFRS 9-style scenario projection logic.
ADV
16
Bayesian Approaches
Prior · Posterior · Beta-Binomial · LDP · Credible Interval
A practical framework for rare-default estimation, prior-supported updating, and the low-default-portfolio logic where classical estimators become unstable.
ADV
track 04
Model Risk, Robustness & Decision Reality
The layer where trust depends on uncertainty, population representativeness, operating thresholds, and explainability.
17
Resampling, Bootstrap & Model Uncertainty
Bootstrap · CI · Cross-Validation · Optimism · Delta Stability
Turns model assessment from a single-number ritual into a stability-focused judgement by adding uncertainty to metrics and challenger comparisons.
MRM
18
Regularization & Shrinkage
Ridge · Lasso · Elastic Net · Bias-Variance · Collinearity
How deliberately adding bias reduces coefficient instability, softens overfitting, and produces models that generalise more reliably.
MRM
19
Segmentation, Reject Inference & Selection Bias
Subgroups · Accept/Reject Bias · Representativeness · Sample Selection
Questions whether the model was actually learned on the right population, especially under filtered samples and shifting business strategy.
MRM
20
Imbalanced Learning, Cutoff Strategy & Cost-Sensitive Decisions
Rare Events · Precision-Recall · Threshold · Expected Cost · Policy
Connects rare-event modelling to approve, reject, escalate, and intervene policy: how the optimal cutoff shifts with imbalance and business cost.
MRM
21
Explainability, Feature Effects & Model Interpretability
Global vs Local · PDP · Contributions · Monotonicity · Governance
Moves from performance to interpretability: how model decisions can be explained, challenged, and governed at both portfolio and individual-case level.
MRM
The intended path is simple: first understand how a model is built, evaluated, and calibrated; then strengthen the statistical foundations behind that logic; then move to time structure and rare-default methods; and finally to model-risk questions on uncertainty, threshold policy, population bias, and explainability.
This page is not just a list of notes. It is a guided route from first contact with modelling toward the harder judgement layer, where a model has to be stable, calibrated, decision-ready, and explainable enough for real use.
FND — Foundations
CORE — Core modelling
ADV — Advanced methods
MRM — Model risk & robustness
Primary references
Definitions and current regulatory statements were checked against official documentation. Library pages describe concepts and APIs; they are not substitutes for model-specific validation.