Skip to content
v0.1.3latest release121 skills4 categories6 sub-categories247 tested scripts397 reference docsContact us
Vector Agentic Skills
Submit a skill

Domain Feature Playbooks

Reference for feature-engineering-idea-generator · 74 lines

AI & Machine Learning / Classical Machine Learningv1.0.0Updated OCT 25 20264 references2 scripts

Starting lists that repeatedly carry signal. Every feature is computed as of the prediction time.

Subscription churn

  • Tenure (days), plan, price paid vs list price, recent price change, contract type, days to renewal.
  • Usage: sessions/logins in 7/30/90 d, trend (7 d vs previous 30 d average), days since last use, core-feature usage, breadth of features used.
  • Support: tickets in 90 d, unresolved tickets, sentiment of last ticket, time to resolution.
  • Billing: failed payments in 90 d, payment method, card expiry within 30 d, discounts ending.
  • Social/network: seats used / seats bought, team members leaving.
  • Leaks to avoid: cancellation reason, retention offer accepted, account status read today.

Card / payment fraud

  • Velocity per card, device, IP, email, merchant: counts and sums in 10 min / 1 h / 24 h / 7 d.
  • Amount vs card's own history (ratio, z-score), round amounts, small "test" transactions before a large one.
  • Novelty: first time this merchant / country / device / shipping address for the card.
  • Geography: distance from home, impossible travel, billing vs shipping mismatch, IP country vs card country.
  • Merchant risk: historical fraud rate of the merchant (out-of-fold), merchant category.
  • Time: hour in local time, night-time flag.
  • Leaks: chargeback flag, analyst decision, case status.

Credit risk

  • Bureau data: utilisation (balance / limit), delinquencies in 12/24 months, inquiries in 6 months, age of oldest account, number of open accounts.
  • Affordability: debt-to-income, payment-to-income, income stability.
  • Behaviour (existing customers): days past due history, payment ratio, cash advances, limit increases.
  • Application: loan amount / income, term, purpose.
  • Regulated: avoid protected attributes and close proxies; monotone constraints for explainability.

Retail demand forecasting

  • Lags: sales at t−7, t−14, t−28, t−364 (horizon-aware: lags must be >= horizon for direct models).
  • Rolling means/medians/std over 7/28/91 days, shifted by the horizon.
  • Calendar: day of week, week of year, month, holidays, paydays, school holidays, events.
  • Price: price, discount depth, price vs 28-day average, promotion flags (planned only).
  • Product/store: category, store size, store-level trends, product age.
  • Stock-outs: flag censored days; do not train on them as zero demand.

Marketing response / conversion

  • Recency of last purchase/visit, frequency, monetary value.
  • Engagement with past campaigns: opens, clicks, conversions per channel.
  • Channel preference, device, time since last contact (fatigue).
  • Product affinity: categories bought, browsing in the last 7 days.
  • For uplift: same features; the treatment must be randomised.

Real estate pricing

  • Size (log sqft), rooms, bathrooms, lot size, age, renovation recency.
  • Location: neighbourhood, distance to centre/transit/schools, local median price per sqft (out-of-fold).
  • Quality/condition grades, amenities (pool, garage), floor level, view.
  • Market timing: month of sale, local price index at sale date.
  • Leaks: final sale price-derived fields, assessed value updated after sale.

IoT / predictive maintenance

  • Sensor statistics over windows: mean, std, min, max, slope, range, skew over 1 h / 24 h / 7 d.
  • Deviations from the machine's own baseline (z-score vs its last 30 days).
  • Frequency-domain features (FFT band energy) for vibration.
  • Operating context: load, temperature, duty cycle, hours since last maintenance, maintenance count.
  • Cross-sensor ratios (temperature per unit load).
  • Leaks: maintenance work orders created after the failure, failure codes.

Healthcare (readmission, risk)

  • Prior utilisation: admissions/ED visits in 6/12 months, length of stay.
  • Diagnoses grouped into chapters or comorbidity indices (Charlson, Elixhauser).
  • Medications count, high-risk medications, lab trends (last value, slope, abnormal flags).
  • Social determinants where allowed.
  • Leaks: discharge disposition coded after the outcome, billing codes finalised later.

B2B lead scoring

  • Firmographics: industry, employee count band, revenue band, region.
  • Engagement: website visits in 7/30 d, pricing-page views, content downloads, email replies.
  • Fit: match to ideal customer profile, technology stack signals.
  • Timing: days since first touch, days since last activity, recent funding or hiring.
  • Leaks: opportunity stage, "sales accepted" flags set after qualification.
More in Classical Machine Learning

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.