EXP-010 / question
Technical question
Can association-rule mining uncover interpretable product co-purchase patterns for retail decisions?
EXP-010 / method
Method and workflow
- Generate synthetic retail transactions with realistic co-purchase bias.
- Mine frequent itemsets using Apriori or FP-Growth through mlxtend.
- Derive association rules and rank them by support, confidence, lift, leverage, and conviction.
- Export item support, frequent itemsets, association rules, summary JSON, and top-item visualizations.
- Interpret discovered rules as retail bundling, merchandising, and recommendation signals.
EXP-010 / skills demonstrated
Data scientist skills shown
Uses frequent itemset mining and association rules to discover patterns rather than predict a target.
Translates product co-occurrence metrics into bundling and recommendation insights.
Exports CSVs, JSON summaries, and figures for inspection and communication.
EXP-010 / evidence
Evidence of work
The README defines support, confidence, lift, leverage, and conviction as outputs for evaluating product relationships.
Both Apriori and FP-Growth are available, demonstrating two classic approaches to frequent itemset mining.
Example rules show product co-purchase relationships such as laptop accessories appearing together more often than random.
EXP-010 / stack
Technical stack
EXP-010 / limitations
Limitations and honesty check
- The transaction data is synthetic, so business conclusions are illustrative rather than production-ready.
- Association rules reveal correlation, not causation.
- Large real-world catalogs would need careful threshold tuning and performance optimization.
EXP-010 / next
Next improvements
- Run the workflow on a real public retail transaction dataset.
- Add rule filtering by category, lift, confidence, and business value.
- Add interactive visualizations for item networks and product bundles.
- Compare rule-based recommendations with collaborative filtering baselines.