A knowledge-based parser: Neural-network-based approaches. Development of a neural network based imputation system for travel diary data
FESSANT ; MIDENET
Type de document
RAPPORT
Langue
anglais
Auteur
FESSANT ; MIDENET
Résumé / Abstract
Work package 5 in the TEST project is dedicated to data validation (erroneous data detection) and data correction (imputation methods). Non responses and erroneous data remain impossible to avoid in survey procedures, whereas their treatment is known to be critical and time and money consuming. Test work package 5 concerns two research teams from FUNDP Namur (b) and from INRETS recoil (f), involved in investigating two different and complementary aspects of this issue. Namur's approach is devoted to improve the automation of incoherence detection and imputation methods thanks to intelligent correction procedures using suitable artificial intelligence computing technologies. In INRETS we have tackled this issue by investigating new statistical methods for imputation. Our initial motivation was to study the feasibility of non conventional neural network based imputation models, both by establishing a state of the art of this particular application field of neural networks, and by experimenting some original models on a concrete and real-size transport survey database. We worked on the trip description database coming from the 1993-94 French national personal travel survey. We also give results concerning the vehicle description files coming from the MEST pilot surveys.