Incident Reporting and Analysis is a sub-area of the field of Accident Analysis. Incident reporting provides an important defence against future failures in many safety-critical industries. Incident Reports provide engineers, managers, operators and others with important guidance about potential problems in existing systems. An incident may be crudely defined as an event that may lead to an accident. There is an empirical relationship of one accident to every 300 incidents. The relative frequency of incidents as opposed to the relative infrequency of accidents, helps to ensure that there is a focus on safety issues [Reason,1998]. While incident reporting systems are of vital importance they often attract little or no attention until a serious accident occurs such as a plane crash, rail crash, or the death of a patient undergoing minor surgery. Following such an event, questions are asked about the causes of the event. In particular it is important to know if this was the first time that the specific causes of the event in question, occurred. If it transpires that similar incidents have indeed previously occurred (even if they did not result in accidents), then the demand for explanations of how this event was allowed to happen again, become a cause of even major concern.

It is now standard practice in safety-critical industries and service organisations (e.g. airlines, railways, hospitals, manufacturing plants etc.) to record detailed reports of incidents and accidents. More and more industries are relying on the insights provided by incident reporting schemes [van der Schaaf,1998]. There is increasing interest in developing large scale international incident reporting systems, for example, EUROCONTROL is pioneering a Europe-wide scheme for incident reporting in Air Traffic Control. The aviation industry was one of the first industries to begin formally recording incidents. The United States ASRS (Aviation Safety Reporting System) was established under the control of NASA in the mid-seventies to collect data on aviation incidents and provide information that could be used to increase air safety. ASRS receives over 2,600 reports per month from pilots, air traffic controllers, cabin crew and others. These reports are analysed and where necessary messages are sent to the appropriate authorities to take corrective actions.

In Ireland, the Safety Department of one airline has over 5,000 incident reports stored in a Incident Management System. This system can provide basic information such as statistics on the number and type of incidents. In addition it can be queried for specific information as the reports are stored in a standard database system. However, the only means of correlating a new (or indeed any) incident with others in the database involves the users composing numerous queries and using their own experience and intuition in deciding what queries might be appropriate. This has been identified as one of the key drawbacks of such passive information systems. The point is that a given incident may be related to one or more previous incidents, in subtle or unexpected ways. Such relationships may not be detected, or even be detectable using standard database queries, hence the need for advanced Incident Management Systems.

It is a non-trivial task to analyse the data gathered by national and international reporting schemes [Johnson,2000]. Specifically, it can be very difficult to identify common causes among the thousands of reports that are received each year. The goal of this research project is to develop techniques to detect the similarity between a new incident and incidents that have already occurred, at the time of incident collection. These techniques could also be used to identify a correlation between incidents already stored in the collection. Such techniques could be used in the development of active incident management systems.

Two techniques will be investigated: classification-based techniques from Information Retrieval and Case-Based Reasoning techniques from Artificial Intelligence. These techniques suggest themselves as a result of the nature of the problem and the nature of incident reports. Incident reports typically consist of a number of pre-specified fields or features in addition to a free text description of the incident. The free text portion of incident reports makes them amenable to processing using IR techniques, however, Johnson reports that standard IR techniques alone are not adequate. On the other hand, the presence of a number of standard features in incident reports makes them amenable to CBR techniques. Feature vectors can be used to represent whether or not particular terms are relevant to a case and there are algorithms for introducing new cases and new features.

Johnson , C.W., Using Case-Based Reasoning to Support the Indexing and Retrieval of Incident Reports, submitted to "Proceedings of European Safety and Reliability Conference (ESREL 2000): Foresight and Precaution", 2000.

Reason, J., Managing the Risks of Organisational Accidents, Ashgate, Aldershot, 1998.

Van der Schaaf, T., PRISMA: A Risk Management Tool Based on Incident Analysis, Proceedings of the International Workshop on Process Safety Management and Inherently Safer Processes, Oct. 8-11, Orlando, Florida, 242-251,1996.