Develop Prediction Model That Can Determine Which Households Are More Likely To Shop and Why

Project Description

InfoScout is the leading provider of real­time shopper insights. Search any brand or retailer to get purchase trends and detailed demographic statistics. ?We collect receipts from a large panel of consumers and have created a massive purchase panel that tracks consumer and household spending longitudinally across many different types of retail outlets and product categories.

Problem we are trying to solve: ?

Using shopper behavior data, household demographics and retailer availability, develop a prediction model that can determine where households are more likely to shop and identify factors that influence their choice of retailer. Identify variables for model development, including those from external data sources ? Perform all data cleaning and manipulation of data set for modeling phases ?

Expertise Required:

Statistics Ph.D. or related ? ML and classification methods (Multi­Class SVM, MNLogit, Bayesian Belief Networks) ? Multiple imputation, sampling methods, factor analysis, PCA,? Distributed processing ?

Our Data Sources:

InfoScout receipt data stored in mySQL database? (~21 Million rows x ~300 columns = ~6.3BN elements) External data source (Experian) format TBD? (Nrows(TBD) x ~3,000 columns)

Our current technology stack:

Vertica v7.1.2­0, MySQL Database, Python, R, Tableau, Microstrategy

Deliverable:

Prediction model that generally explains consumer behavior across a number of inputs (transactions, geo­demographics, other attitudinal data), iteration and refinement, ? Model simulator, ? Summary report of findings,  Model Specs and Diagnostics