2026-10-07T19:11:43Z https://api.figshare.com/v2/oai

oai:figshare.com:article/33104105 2026-10-02T14:23:47Z category_26971 category_26557 category_29173 item_type_3 month_year_10_2026

On-Road Vehicular Emission Measurement Data & Analysis leakage codes Mohd Kamil Vakil (24474350) Pollution and contamination not elsewhere classified Air pollution modelling and control Machine learning not elsewhere classified cross-validation leakage random forest vehicle emissions leave-one-group-out small-fleet machine learning <p dir="ltr">Random-forest and other flexible regressors fitted to small-fleet vehicle emission datasets often fail to generalize when validated on held-out vehicles, yet appear accurate under standard random cross-validation. We demonstrate this on a nine-vehicle Indian in-use fleet, showing that adding static vehicle attributes (kerb weight, engine displacement, odometer) raises random-fold R² to 0.60–0.86, but leave-one-vehicle-out R² drops to −0.47 to −1.64. The mechanism: the three attributes jointly identify every vehicle uniquely, so the model functions as a vehicle-identity lookup rather than an emission model; random splitting leaks this identity across train/test boundaries. A decision-tree classifier recovers vehicle identity from these attributes with 100% accuracy. Pooling more training vehicles (2–8) does not close the gap. We recommend leave-one-vehicle-out or leave-one-group-out validation as a minimum reporting standard for any fleet-emission model including vehicle-constant features, and quantify on real data the error that random-split reporting alone would produce.</p> 2026-10-02T14:23:47Z Dataset Dataset 10.6084/m9.figshare.33104105.v2 https://figshare.com/articles/dataset/On-Road_Vehicular_Emission_Measurement_Data/33104105 CC BY 4.0