The Paari School of Business at SRM University-AP is organising a Research Seminar titled “Managing Density Bias In Clustering-Based Facility Placement For Consumption-Driven Last-Mile Delivery: A Demand-Conditioned Clustering Approach”. The session will feature Dr Prithvirajan D, PhD alumnus from the Indian Institute of Science (IISc) Bengaluru, as the expert speaker.
Dr Prithvirajan D received his PhD from the Department of Management Studies at IISc Bengaluru, specialising in decision support systems, artificial intelligence, operations management, and supply chain optimization. He actively pursues entrepreneurial initiatives translating research outcomes into industry solutions, notably developing “Fuel Fast”, a production-grade decision support system for LPG cylinder last-mile delivery operations in India.
The session is expected to benefit students, researchers, and faculty members interested in supply chain optimization, operations research, machine learning applications in logistics, and business analytics.
Abstract
Consumption-driven last-mile delivery systems exhibit intermittent customer ordering behaviour, where only a subset of registered customers generates demand within a planning period. In such settings, clustering-based planning approaches that determine temporary restocking or consolidation locations using the geographic distribution of all registered customers can misalign facility locations with realised demand. This occurs because clustering algorithms tend to position centroids toward dense customer regions irrespective of customer activity. Under intermittent demand, this density-seeking tendency creates an operational misalignment between facility locations and realised demand, referred to in this study as density bias, which needs to be managed rather than eliminated.
This study examines density bias in conventional clustering-based facility planning and proposes a demand-conditioned clustering approach that incorporates anticipated demand into clustering to better align facility locations with expected demand. Following a predict-then-optimise paradigm, a machine learning-based classification model first identifies customers likely to generate demand within the planning period, after which temporary restocking locations are determined by clustering only the anticipated active customers.
The proposed approach is evaluated using a large-scale real-world LPG distribution system comprising 16,861 customers and 641,276 historical orders. Experimental results across multiple population distribution scenarios show that demand-conditioned clustering reduces total travel distance by 36-52% compared with current practice and by 9-18% relative to conventional clustering-based planning. These findings demonstrate that incorporating demand anticipation at the clustering stage can manage density bias, reduce avoidable replenishment travel, and improve operational efficiency in consumption-driven last-mile delivery systems.
Key Details
- Date: October 21, 2026 (Wednesday)
- Time: 03:00 PM to 04:00 PM
- Venue: Level 5, Tiered Classroom, Homi J Bhabha Block
- Topic: Managing Density Bias In Clustering-Based Facility Placement For Consumption-Driven Last-Mile Delivery: A Demand-Conditioned Clustering Approach
- Speaker: Dr Prithvirajan D (PhD, Department of Management Studies, IISc Bengaluru)