Rules and models together
| Component | Role |
|---|---|
| Hard rules | regulatory blocks, sanctions, known compromised cards, impossible values |
| Model score | ranks the remaining risk |
| Score bands | approve (low), step-up authentication (medium), analyst review (high), decline (very high) |
| Analysts | label the uncertain band and feed decisions back |
| Rule engine for new attacks | fast, temporary response until the model is retrained |
Rules are explainable and fast to deploy, but they decay and overlap. Review the rule hit rates monthly and retire rules that the model already covers.
Feedback loops and selection bias
- Declined transactions never produce chargebacks, so the model never learns whether they were fraud. Keep a small random sample of high-score transactions that are reviewed instead of blocked, or use analyst outcomes as labels.
- Analyst decisions are labels with their own bias and delay. Record the decision, the reason and the time.
- When comparing a new model with the incumbent, both must be evaluated on traffic the incumbent did not filter, or the comparison favours the incumbent.
Adversarial drift
Fraudsters adapt within days or weeks:
- retrain frequently (weekly or monthly) with an out-of-time validation for each release
- monitor AP by week, the alert volume, the share of fraud caught by rules vs the model, and feature PSI
- run anomaly detectors next to the supervised model to catch new patterns without labels
- prefer features that are costly for fraudsters to fake (device, network, behaviour) over easily changed ones
Real-time architecture
event -> feature service (online counters, profile store) -> model (< 50 ms) -> decision engine (rules + bands)
-> response (approve / challenge / decline) -> logging (features, score, decision, latency)
-> label pipeline (chargebacks, analyst outcomes) -> training data (point-in-time joins)Log the exact features used at decision time. Training on recomputed features introduces skew.
Monitoring checklist
- average precision and value detection by week (with label maturity)
- alert volume per day vs capacity; analyst hit rate
- false-positive complaints and step-up abandonment rate
- feature PSI and missing rates; online/offline feature parity checks
- latency and timeouts (a timeout must fall back to a safe default)
- fairness: decline and challenge rates by customer segment and region
Governance
Document the label definition, features, thresholds and their cost rationale, the expected performance and known blind spots in a model card (model-card-writer). Credit and payment fraud decisions can trigger adverse-action or explanation obligations. Keep the reason codes and the review trail.
anomaly-detection-model-builder
Build and evaluate anomaly detection for tabular, sensor and transaction data with few or no labels: define the anomaly type (point, contextual, collective), choose detectors (robust z-score/MAD and IQR rules, Mahalanobis with robust covariance, k-NN distance, LOF, Isolation Forest, one-class SVM, PCA reconstruction / autoencoders, forecasting residuals for time series), combine them by rank, explain each alert by its deviating features, set thresholds from an alert budget or the score distribution, and evaluate with precision@k, PR-AUC and recall on known incidents, or detector agreement when no labels exist. Includes pure-Python detectors (robust z, Mahalanobis, kNN, Isolation Forest, ensemble) and an evaluation script.
catboost-model-builder
Build CatBoost models that use its strengths correctly: raw categorical columns as cat_features (ordered target statistics, no manual encoding), text_features, Pool construction with NaN-safe categories, ordered boosting and has_time for temporal data, symmetric (oblivious) trees, overfitting detector with eval_set and use_best_model, losses for classification, regression (Quantile, Poisson, Tweedie) and ranking (YetiRank), class weights, monotone constraints, feature importance (PredictionValuesChange, LossFunctionChange) and SHAP values, GPU training and model export. Plans cat/text/numeric columns from a CSV and generates a runnable training script, and checks parameter sets for the conflicts CatBoost rejects (iterations vs n_estimators, Bayesian bootstrap with subsample, multiple class-weight options, depth > 16, use_best_model without eval_set).
categorical-encoding-selector
Choose and implement the right encoding for every categorical column given the model family and the data: one-hot (with infrequent-level grouping and handle_unknown), ordinal with an explicit order, cross-fitted target encoding (scikit-learn TargetEncoder, CatBoost ordered statistics, category_encoders inside CV), frequency/count encoding, feature hashing, native categorical support in LightGBM, HistGradientBoosting, XGBoost and CatBoost, prefix levels for codes (zip3, NAICS-2), and embeddings. Measures cardinality, rare-level share, unseen-level rate on a time-ordered holdout and target signal per column, emits the ColumnTransformer, and quantifies target-encoding leakage (in-sample vs out-of-fold vs holdout AUC).
classification-model-selector
Choose the right classification algorithm for tabular, text or mixed data and prove the choice with a fair bake-off: logistic regression (L2, L1, elastic net), linear SVM, naive Bayes, k-NN, RBF SVM, decision trees, random forest, extra trees, HistGradientBoosting, LightGBM, XGBoost, CatBoost, explainable boosting machines (EBM/GAM), points scorecards and MLPs, matched to data size, width, categorical and high-cardinality share, missingness, class imbalance, text, and constraints such as interpretability, monotonic constraints, calibrated probabilities, latency budgets and tiny deployable models. Ranks candidates with starting hyperparameters and cautions, then generates a bake-off script running every candidate on identical CV folds with per-fold metrics, fit time and prediction latency.
clustering-algorithm-selector
Choose the right clustering algorithm and check that clustering makes sense at all: cluster tendency (Hopkins statistic with a correlated-features caveat, nearest-neighbour isolation), then k-means / MiniBatchKMeans, Gaussian mixtures with BIC, agglomerative (Ward, average, complete, single) with dendrograms, DBSCAN, HDBSCAN, OPTICS, spectral clustering, BIRCH, mean shift, k-prototypes and k-modes for mixed or categorical data, and Gower distance, matched to data size, expected shapes, noise, varying density, known k, soft memberships, hierarchy and streaming. Covers preprocessing (scaling, encoding, dimensionality, distance metrics), validation (internal indices, bootstrap stability, external labels, business profiling) and assigning new points.
collaborative-filtering-implementer
Implement collaborative filtering recommenders that beat popularity: item-based kNN with cosine, Jaccard or asymmetric similarity, shrinkage, neighbourhood size, positive-only or mean-centred rating signals and 'because you liked X' explanations; implicit ALS with confidence weighting, BPR pairwise ranking and the closed-form EASE model in plain NumPy, tuned on a temporal validation split by recall@k and NDCG@k and refit on all data; plus user-kNN, explicit matrix factorisation, the implicit/LightFM libraries, hyperparameter ranges, cold-start fallbacks, scaling with sparse matrices and approximate nearest neighbours. Produces top-k lists in the format of the recommendation evaluator.