Vehicle reidentification is the process of tracking a vehicle along a highway as it crosses detection stations. Inductive loop detectors are by far the most widely deployed vehicle detectors. In the present work, vehicle reidentification is performed by combining vehicle-specific information (length and electromagnetic signatures) and some contextual information (lane, speed, and time) to form a decision tree. This approach provides a specific decision tree for tracking vehicles along each highway section. After training, the decision tree successfully classified about 95% of the unseen test records—a significant improvement relative to the literature and our own previous work on the same data. This success rate has been consistently obtained from two data sets: one consisting only of passenger vehicles and another consisting of a representative traffic mix.
A framework for representing and reasoning about uncertain temporal information is presented. This framework relies on a non-recursive representation of probabilities as functions of time. The representation is compatible with survival analysis techniques, and allows the use of regressive survival models to represent common interaction patterns. In another departure from previous temporal probabilistic knowledge representations, surprise measures are used to detect and correct poor probability estimates. The present framework relies mainly on a theory of causality, while still allowing the representation of phenomena without requiring a complete understanding of causes.
The present paper reports experimental densities and kinematic viscosities of ten binary mixtures of 1-hexanol, ethylbenzene, octane, p-xylene, and chlorobenzene at 293.15 K and 298.15 K. The experimental data were obtained using Cannon-Ubbelohde viscometers and Anton-Paar DMA 60 density meter under temperature control. The experimental data were correlated with five different models: Allan-Teja, GC-UNIMOD, the Generalized Corresponding States Principle (GCSP), generalized McAllister three-body interaction, and an Artificial Neural Network (ANN) viscosity model. Among the tested models, the generalized McAllister three-body interaction model yielded the most accurate predictions, with an overall average absolute deviation of 3.9%. The Allan-Teja correlation exhibited the largest deviations for binary mixtures containing chlorobenzene. Two specialized artificial neural networks (ANN) models were developed to predict kinematic viscosities at 293.15 K and 298.15 K: one trained on 1-hexanol-containing mixtures and the other on non-associating hydrocarbon mixtures. Each model employed a 6–4–1 feedforward architecture with backpropagation training. The class-specific ANN models achieved average %AAD values of 3.3% and 1.6%, respectively. However, these models are limited to the mixture classes on which they were trained and cannot be generalized to other systems.