Document Type
Student Coursework
Publication Date
Fall 10-30-2026
Abstract
Historical order archives and live transaction streams support essential, complementary operational capabilities in online grocery retailing. However, unifying them requires consistent dimensional modeling, deterministic deduplication, and rigorous batch-to-stream reconciliation. This study presents the design, implementation, and empirical evaluation of an integrated big data architecture combining PySpark batch processing, exact sparse-matrix co-occurrence projection, distributed FP-Growth, gradient-boosted reorder classification (XGBoost), and event-time Structured Streaming. Benchmarked on the complete 33.8-million-row Instacart dataset (3,421,083 orders across 206,209 consumers), the batch workflow cleaned, validated, and joined all records in 3.9 minutes without data loss. Market basket mining over 3,346,083 observed baskets identified 528 association rules at 0.2% support and 10% confidence. Peak lift concentrated within intra-brand multi-flavor clusters (lift up to 73.6) rather than traditional staples, while cross-aisle culinary pairings consistently yielded actionable affinities (lift ≥ 5.8). Reorder forecasting on a customer-stratified hold-out split (210,872 pairs) achieved an ROC AUC of 0.83 (F1 = 0.44), where consumer purchase recency and individual product interaction frequency accounted for 61.8% of total feature importance gain. Under adversarial streaming replay with injected redeliveries, late arrivals, and corrupted foreign keys, stateful event-time watermarking and quarantine routing produced an exact 100% reconciliation match against offline ground truth (27,465 orders; 324,026 lines). Finally, we formalize a cloud deployment topology on Google Cloud Platform, providing an empirical baseline for production-grade grocery telemetry.
Program or Discipline Name
Computer and Information Sciences
Recommended Citation
Kumar, K., & Cha, S. (2026). Evaluating Scalable Market Basket Analytics and Reorder Prediction in a Unified Batch-and-Streaming Architecture. Retrieved from https://digitalcommons.harrisburgu.edu/other-works/38
Start Page No.
1
End Page No.
8
Creative Commons License
This work is licensed under a Creative Commons Attribution 4.0 International License.
Submission aggreement