TY  - CONF
TI  - Data-driven decision support system for resource allocation to cast in-situ piles of linear projects
AU  - Anantharaam, Ramachandran
AD  - Master of Technology, Department of Civil Engineering, Indian Institute of Technology Madras, Chennai, India,  ananthu_357@alumni.iitm.ac.in, orcid.org/0009-0007-8503-1867
ED  - Hamzeh, Farook
ED  - Poshdar, Mani
ED  - Garcia-Lopez,, Nelly P.
ED  - Gan, Vincent
PY  - 2026
DA  - 2026/06/22
T2  - Proceedings of the 34th Annual Conference of the International Group for Lean Construction (IGLC 34)
C3  - Proceedings of the 34th Annual Conference of the International Group for Lean Construction (IGLC 34)
CY  - Singapore, Singapore
SP  - 1713
EP  - 1724
DO  - 10.24928/2026/0268
SN  - 2789-0015 (ISSN)
AB  - Field observations from an active cast-in-situ piling site revealed that resource-related waiting contributed to 54% of total operation lead time, disrupting workflow and production reliability. This study investigates whether near-term resource requirements in piling operations can be predicted using process-state information available during execution. A data-driven framework is developed to predict demand for key piling resources (water, tremie pipe, rebar cage, casing, etc.) within a 30-minute lookahead window. Time-stamped operational data was collected from 13 piles, capturing features such as pile geometry, current activity state, depth progression rate, subsurface conditions and recent execution disruptions. Resource demand was framed as a binary classification problem for each resource type. Random Forest classifiersI were trained on 80% of the data and validated with the rest. The models achieved strong performance: F1-scores of 0.90 (water), 0.97 (tremie pipe), 0.95 (rebar cage) and so on. Deploying these predictors reduced resource-induced waiting to 12% and increased Kelly efficiency by 16%. This work operationalizes predictive resource planning at the activity level, advancing lean construction practice by enhancing flow reliability. Future research could integrate these predictors into Last Planner System workflows, couple them with discrete-event simulation or digital twins for system-level impact assessment, and generalize to other resource-sensitive construction activities.
KW  - Resource prediction
KW  - make-ready planning
KW  - waiting time
KW  - production planning
KW  - machine learning for construction.
L1  - https://www.iglc.net/papers/details/2556/pdf
UR  - https://www.iglc.net/papers/details/2556
DB  - IGLC.net
LA  - English
N1  - Export date: 04 October 2026
ER  -