Tensorial Principal Component Analysis in Detecting Temporal Trajectories of Purchase Patterns in Loyalty Card Data:...

Area of Science:

• Consumer behavior analysis
• Data science and analytics
• Public health research

Background:

• Retail loyalty card data offer a rich source for analyzing customer health-related purchasing habits.
• Customer purchase data, including expenditures and timing, can be structured as 3D tensorial data.

Purpose of the Study:

• To apply tensorial principal component analysis (PCA) to loyalty card data for uncovering health-related purchase patterns.
• To identify customer segments with unique purchasing behaviors.
• To compare the utility of tensorial PCA with standard PCA for this application.

Main Methods:

• Utilized loyalty card data from 7251 Finnish retail customers from 2016.
• Reclassified purchases into 55 product groups and aggregated data across 52 weeks.
• Applied tensorial PCA to simultaneously reduce time and product group dimensions, using augmentation for principal component selection.

Main Results:

• Tensorial PCA effectively identified typical food purchasing patterns across time and product categories.
• Distinct purchasing behaviors among customer groups were detected, exemplified by meat product consumption patterns (stable, increasing, decreasing, or seasonal).

Conclusions:

• Tensorial PCA provides a more detailed examination of purchasing behavior than traditional methods by handling time and product dimensions concurrently.
• Future research can link identified patterns to socioeconomic factors and external influences to guide consumers toward healthier food choices.