Brainstorm and prioritize features for a predictive model
A ranked list of candidate features grounded in the problem, with leakage checks, build effort, and expected signal.
The prompt
Help me brainstorm and prioritize features for a predictive model.
What the model predicts: [target definition, unit, prediction time, outcome window]
Available data sources: [PASTE table names and key columns]
Model type planned: [e.g. gradient boosting, logistic regression, not decided]
Current features (if any) and performance: [LIST, with metric]
Steps:
1. Generate candidate features grouped by theme (e.g. recency, frequency, monetary value, engagement trends, tenure, product usage, support interactions, seasonality, external factors). For each, write the exact calculation and time window.
2. Include change-over-time features (last 7 days vs. prior 30), ratios, and counts of events, not just raw values.
3. Leakage check: flag any feature that uses information not available at prediction time, or that is a proxy for the target.
4. Fairness check: flag features that are, or proxy for, protected characteristics, and note whether they should be excluded.
5. Rate each feature: expected signal (high / medium / low, with reasoning), build effort, and data quality risk.
Return:
- A feature table: Feature | Calculation | Window | Source | Expected signal | Effort | Leakage risk | Notes.
- The top 10 to build first.
- Encoding and handling notes: missing values, high-cardinality categories, outliers, scaling if needed.
- How to test whether new features actually help (ablation, feature importance with caution, time-based validation).
Highlighted [brackets] are blanks to replace with your own details.
Fill in the blanks and use this promptMore Data & Analysis prompts
What to fill in
- target definition, unit, prediction time, outcome window
- PASTE table names and key columns
- e.g. gradient boosting, logistic regression, not decided
- LIST, with metric
How this prompt is built
This prompt follows the Few-Shot structure (Few-shot prompting). A few-shot prompt includes one or more worked examples of the input and the output you want, so the model can copy the pattern instead of guessing it.
Related prompts
- Write an Excel formula to categorize records automaticallyBuilds the formula and lookup table that auto-assigns categories, bands, or tiers to each row based on your rules.
- Write an Excel formula to extract text, numbers, or dates from messy cellsPulls the piece you need out of combined or messy cells, like codes, names, domains, or dates, with a formula that handles exceptions.
- Build a best, base, and worst case scenario model in ExcelTurns your assumptions into a simple Excel scenario model with an inputs tab, a scenario switch, and side-by-side results.
- Find recurring patterns across incidents, complaints, or ticketsAnalyzes a pasted list of incidents, complaints, or tickets to find the patterns, clusters, and repeat causes behind them.