Use of ILP to investigate accident data (OUCL, Smith System Engineering)

Application domain: Investigating road traffic accidents
Source: Transport Research Laboratory
Dataset Size: 339KB (tar.gz, including LATEX documentation)
Data Format: Prolog
Systems used: CProgol4.2
References: (Roberts et al. 1998)

The data (OUCL, Smith System Engineering)

The data consists of Prolog translations of two databases. The first is the STATS19 database, which records factual details of road traffic accidents such as road surface conditions, weather details, casualties, and vehicle manoeuvring details. The second is the accident causation database, which records more subjective information concerning the accident such as who caused the accident, what were the underlying causes and what precipitated the accident. Accidents are referenced using a reference number common to both datasets, which include a total of 1409 accident reports.

Experiments with CProgol4.2 (OUCL, Smith System Engineering)

Investigating young male drivers

Young male drivers have many more accidents per mile than do other drivers. The purpose of this investigation was to investigate if they were overrepresented in a particular type of accident, or if they just have more accidents. The latter conclusion was supported; young males appear to have accidents due to inexperience, without any mitigating circumstances. Further details including the results of this investigation have been published in (Roberts et al. 1998).

Experiments with Midos (GMD)

Following the transport workshop held at York, Midos was used to analyze the traffic data, using the entire dataset of 1409 accident report to screen for interesting subgroups of accidents involving young male drivers. Several interesting subgroups were discovered and will be compared with the rules discovered by Progol.


  1. S. Roberts, W. Van Laer, N. Jacobs, S. Muggleton, and J. Broughton. A Comparison of ILP and Propositional Systems on Propositional Traffic Data. ILP-1998: 291-300.

back to index