Customer Intelligence Pipeline
AAI 5025 final project
A course pipeline that turns SQLite customer data into cleaned features, spending models, and customer segments.
IllustrationHighlights
- SQLite extraction through feature engineering (spend-per-visit, TextBlob sentiment) to modeling
- Predicted customer spend at R-squared 0.76 with Linear Regression, on 200 held-out rows
- Produced three K-Means customer segments (high income/high spend, low/low, low income/high spend)
How it was built
I built this course capstone as one notebook. It extracts customer records from SQLite, removes two duplicate rows from 1,002, imputes missing education values and engineers features such as spend per visit and TextBlob sentiment. A linear regression predicts total spend at R-squared 0.76 on 200 held-out rows, and K-Means with three clusters gives customer segments. I scaled using the training split only and mapped centroids back to dollars for interpretation. A nearest-neighbour model for gender scored 0.455, below its 0.509 majority baseline, and the notebook reports that as a negative result. Limits: single course dataset.