Empirical Evaluation of Maximum Entropy OD Matrix Estimation Under Limited Traffic Count Data for Andalalin Studies

Abstract

OD matrix estimation under limited traffic count data in Indonesian Traffic Impact Analysis (Andalalin) remains fundamentally underdetermined, as the number of unknown OD pairs far exceeds the available traffic count observations. This study develops and empirically evaluates an integrated OD estimation system combining gravity model calibration, maximum entropy (MaxEnt) updating, and user equilibrium assignment within a Design Science Research framework. A stratified-coverage subsampling procedure (31 runs across four data availability scenarios) was applied to two datasets with contrasting characteristics. On the Magelang benchmark dataset (7 zones, known ground-truth OD), three key findings emerged: (1) reducing observed links from 29 to 18 via representative stratified selection improved GEH pass rate from 37.9% to 62.4%, indicating an optimal constraint count for small underdetermined networks; (2) GEH improvement did not coincide with improved OD accuracy (RMSE_OD increased 49%), empirically confirming the inherent limitation of flow-fit metrics as OD quality proxies; (3) result variability increased sharply below ~15 observed links, indicating a data threshold for reliable estimation. On the RSU Tabanan case study (4 zones, field data), all 31 runs produced identical outputs regardless of subset selection, revealing a prior-dominated regime caused by a 2.57× demand-volume structural inconsistency. These findings constitute a two-regime diagnostic framework applicable to Andalalin survey planning.