INTEGRATION OF APRIORI ALGORITHMS WITH

CASE-BASED REASONING FOR FLIGHT ACCIDENT

INVESTIGATION

Nan-Hsing Chiu

Department of Information Management, Ching Yun University, Jhongli city, Taoyuan, Taiwan, R.O.C.

Pei-Da Lin

Occurrence Investigation Division, Aviation Safety Council, Taipei, Taiwan, R.O.C.

Chang En Pu

Investigation Bureau of MOJ, Taipei, Taiwan, R.O.C.

Keywords: Accident Investigation, Decision Support Systems, Case-based Reasoning, Apriori.

Abstract: The analysis of flight accidents has been demonstrated to be a crucial tool for improving flight safety. The

utilization of visual decision support systems potentially assists investigators in quickly and accurately

identifying the underlying causes of accidents. This study aims at supplying a visual decision support

system, based on the Apriori and case-based reasoning approaches, for assisting investigators in analyzing

human injuries in flight accidents. We demonstrate our approach using the aircraft configuration of flight

CI611. The experimental results show that the proposed approach provides support for quick decisions by

investigators on the basis of a visualization system.

1 INTRODUCTION

With modern aviation technology, airplanes have

become one of the safest means of transportation.

There were about 21,380 flight incidents and

accidents between 1989 and 1999 (Weng et al.,

2003). It is important in any flight accident to ensure

that human injuries are as few as possible. The

investigation of aircraft accidents is focused on the

circumstances of the accidents, including analyzing

all available information in order to draw

conclusions. The objective of the investigation of

flight accidents is to enhance air traffic safety by

introducing recommendations intended to prevent

accidents in the future.

An investigation of an aircraft accident engages a

variety of personnel: specialists, experts, legal

authorities, and accredited representatives, including

representatives for agents of design, manufacture,

operation, and so on (Milosovski et al., 2008).

Investigators need to pursue a comprehensive

examination of the accident site, wreckage, witness

information, any recorded media, component

examinations, tests, simulations, and other evidence

in order to determine the cause of an accident. This

shows how an aircraft accident investigation is a

complex issue involving many different factors.

A detailed analysis of flight circumstances is an

important basis of investigation and is essential in

identifying the underlying factors leading to

accidents. The Apriori algorithm is a well-known

association rule approach that is able to derive

frequent itemsets from a variety of datasets (Cristian

& Mitica, 2007). The case-based reasoning (CBR)

approach also demonstrates high reliability on

similar measurements among diverse data sources

(Chiu & Huang, 2007). Therefore, this study aims to

combine these two approaches for visually

investigating and analyzing cases of human injury

during flight accidents.

487

Chiu N., Lin P. and Pu C. (2010).

INTEGRATION OF APRIORI ALGORITHMS WITH CASE-BASED REASONING FOR FLIGHT ACCIDENT INVESTIGATION.

In Proceedings of the 5th International Conference on Software and Data Technologies, pages 487-490

DOI: 10.5220/0002922204870490

 SciTePress

2 LITERATURE REVIEW

The improvement of flight safety involves attending

not only to factors that increase the likelihood of

crashes occurring but also to factors that increase the

likelihood of both fatal and non-fatal injury in the

crashes that do occur. In 2003, O'Hare et al. (2003)

utilize integrated database information to identify

risk factors for both fatal and non-fatal outcomes for

all civil aviation crashes. Their case study shows that

the most significant factors were post-crash fire, off-

airport crash location, aerobatic flight, lack of an

airworthiness certificate, and aircraft classified as

microlight. Taneja & Wiegmann (2003) performed a

descriptive study of fatal injuries based on autopsy

reports. Their results demonstrate that the most

significant factors of fatal injury are post-crash fire

and lack of shoulder harness use. Each flight

accident usually involves unique reasons for fatal

and non-fatal outcomes. Given the diversity of

aircraft and autopsy reports, a user-friendly tool

would play an important role for investigators to be

able quickly to identify the key factors in human

injury in flight accidents.

Providing investigators with a summary of

information from an accident’s injury data plays a

crucial role in determining the probable causes of

injury. Agrawal et al. (1993) introduce mining

association rules based on the Apriori algorithm.

This algorithm finds frequent itemsets and is the

most popular algorithm for association rules

discovery. Chiu et al. (2006) utilized the benefits of

an approach using the Apriori algorithm for

identifying the malfunction of electronic ballasts in

an aircraft. They compared the Apriori algorithm

with a simple genetic algorithm. Their experimental

results demonstrate that their approach achieves

higher accuracy in distinguishing and classifying

malfunctions in aircraft.

Because human injuries with the exact same type

are usually rare occurences within the overall

dataset, providing investigators with a visualization

of similar cases is a useful approach for

understanding frequent itemsets and their

distribution in a given aircraft accident. CBR is a

data-intensive method that matches an input pattern

with all the information in the database. It searches

the repository for all existing datasets that have

attributes similar to those of the new dataset and

retrieves the nearest measurement value as

estimates. Paul et al. (2001) present a CBR approach

for quantifying confidence in the probability that a

program is free of specific classes of defects. Their

results show that it is possible to analyze Y2K

defects based on defect data. Chiu & Huang (2007)

investigate the effect of CBR on the improvement of

estimation accuracy. Their empirical results show

that their proposal is a feasible approach for

improving the estimation abilities.

CBR is a well-known approach for solving

problems on the basis of similar measurements.

Thus, for quickly identifying similar cases for

investigators, , this study utilizes an approach based

on the visualization of aircraft configuration in order

to illustrate integrated information. This integrated

information is derived by combining the association

rule approach of the Apriori algorithm with the

similar measurements used in CBR for further

improvement in the aircraft accident investigation

process.

3 METHODOLOGY

Figure1 shows the framework for visual analysis of

human injuries in flight accident investigations and

is based on the integration of the Apriori and CBR

(ACBR) approaches. The flight accident database

includes three data sources: a passenger database, a

family member database and an autopsy report

database. The passenger database includes all of the

basic data for the flight crews and passengers. The

family member database includes deoxyribonucleic

acid (DNA) data for family members of the flight

crews and passengers in order to match DNA. The

autopsy reports show DNA and relative information

about the bodies of the dead. They also describe

relative information for passengers with slight or

serious injuries. The DNA matching stage ensures

correct connections among passenger database

entries, family member database entries and autopsy

reports.

Autopsy

Reports

FamilyMembers

Database

DNAMatch

Apriori

FlightAccident

Database

Aircraft

Configuration

Retrieve SimilarityMeasure Reuse

VisualAnalysis

CBR

Passenger

Database

Figure 1: A framework for ACBR.

ICSOFT 2010 - 5th International Conference on Software and Data Technologies

488

The Apriori algorithm is adopted for deriving

frequent itemsets from the flight accident database

as cases for analysis. Using the Apriori algorithm,

human injuries with categorical attributes are used to

construct association rules for determining the

frequent itemsets. Association rules are statements

of the form “if antecedent(s), then consequent(s).” It

is an implication of the form a→b, where a is the

conjunction of conditions and b is the result of their

association. Support refers to the percentage of

records in the training data for which the antecedents

(the "if" part of the rule) are true. Confidence is

based on the records for which the rule's antecedents

are true and is the percentage of those records for

which the consequent(s) are also true. Given a set of

human injury datasets, we wish to generate all the

association rules that have greater support and

confidence than the user-specified minimum support

and confidence.

The CBR approach for visual analysis of human

injuries includes three stages: retrieval, similarity

measure and reuse. For a case under analysis, this

approach retrieves the human injury cases from the

flight accident database. The distance between the

case under analysis and other human injury cases in

the flight accident database is measured. Among all

human injury cases, the most similar one will show

the lowest distance from the case under analysis. As

each human injury case has a respective distance

indicating similarity, the visual analysis system plots

similar cases using different colors on the aircraft

configuration image. On the basis of visual

recognition, investigators can easily and quickly

understand similar cases from a large number of

diverse data sources.

The similarity measure of the CBR approach is

shown in Equation (1), where C

is the case being

clustered and C

is the human injury case from the

flight accident database. For each case being

clustered, there are n feature values in conjunction

with feature weight w. Investigators are able to

assign feature weights for different features,

depending on their requirements. Using the values of

features normalized between 0 and 1, the similarity

measurement determines the straight-line distance

between two features of a human injury case. The

distance decreases as similarity increases. The sum

of the squares of the distance for each feature is the

square of the distance between two human injury

cases. Therefore, the closest human injury case to

the case C

is the case with the minimum distance.







hfiffhi

CCWCC

)(),(



(1)

4 EXPERIMENTS

An aircraft configuration matching flight number

611 (CI611) from China Airlines was adopted to

construct our system. The aircraft was a Boeing 747-

200 on a regularly scheduled flight from Chiang Kai

Shek International Airport (now Taiwan Taoyuan

International Airport) in Taiwan to Chek Lap Kok

International Airport in the Hong Kong special

administrative region. This aircraft broke into pieces

in mid-air and crashed while in flight on 25 May

2002. All 225 people aboard the flight were killed in

this accident.

Given the right to privacy of flight crews and

passengers, the simulated data of 225 cases are used

to demonstrate the visualization system. Every

human body is divided into 14 sections for analysis.

In addition to the sections of head and body, the left-

hand side and right-hand side of the body are

accounted for in six parts consisting of the arm,

elbow, palm of the hand, thigh, shank and palm of

the leg. In the association results from the Apriori

analysis of this dataset, the most frequent itemset is

the head and body. There are eleven cases with

frequent injuries to the head and body in seats from

25D to 25G, 26E to 26G and 27D to 27G.

For the proposed model, or ACBR approach, the

user interface for assigning features and weights to

different sections of human body is shown in Figure

2. In addition to the sections of the head and body,

this user interface provides the 12 options described

above for retrieving similar cases from the relational

databases. The two sections of the right arm and left

arm are near to the head and body and are selected

with lower weights than are the head and body. Two

other sections, the right elbow and left elbow, are

also assigned lower weights than the previous four

parts, sections 1, 2, 3 and 6. These weights are

determined based on the requirements for analysis

and the knowledge of the investigators.

Figure 2: User interface for feature weights.

INTEGRATION OF APRIORI ALGORITHMS WITH CASE-BASED REASONING FOR FLIGHT ACCIDENT

INVESTIGATION

489

Figure 3: The visulization results of ACBR.

Figure 3 shows the visualization results of the the

most frequent itemset used to measure similarities to

all the other cases of the ACBR approach. There are

108 cases showing lower similarities to higher

similarities to the case of the most frequent itemset

involving head and body injuries. This visual system

shows the relative similarity between a particular set

of cases and all the other cases from the most similar

to the least similar. It also reveals the group from

seat rows 49 to 52 who had similar injuries, which

provides further directions for investigators to

explore the causes of injury.

5 CONCLUSIONS

An investigative result of a flight accident cannot

reveal corresponding accidents if the relative

information is inadequate. The ability to demonstrate

visually the various data could be very helpful for

investigators in understanding the relevant

information. The present paper proposes an

improvement of detailed analysis of accident data

using the integration of the Apriori and CBR

approaches for visually analyzing flight accidents.

The frequent itemsets representing human injuries

are identified using the Apriori approach, and the

similar injury cases are confirmed based on the CBR

approach.

The experimental results are encouraging when

the association rules of the Apriori and CBR

approaches are both used in the visualization system

using the aircraft configuration and accident data

from flight CI611. The frequent itemsets are quickly

retrieved using the Apriori approach, and it provides

objective itemsets, rather than human injuries that

are subjectively identified by experts or

investigators. We are encouraged by the results of

the present study and are interested in investigating

whether the use of different configurations (e.g.,

fuzzy logic or grey relational analysis) would result

in further enhancements of this method in the future.

ACKNOWLEDGEMENTS

This research was supported by National Science

Council, Taiwan, Republic of China, under the

contract number NSC 98-2410-H-231-006-, NSC-

97-3114-P-707-001-Y and NSC 98-3114-Y-707-

001.

REFERENCES

Agrawal, R., Imielinski, T., and Swami, A., 1993. Mining

association between sets of items in massive

databases, In Proceedings of the ACM SIGMOD

International Conference on Management of Data. pp.

207-216.

Cristian, A., Mitica, C., 2007. Grid implementation of the

Apriori algorithm, Advances in Engineering Software.

Vol. 38, pp. 295-300.

Chiu, C., Hsu, P. L., and Chiu, N. H., 2006. Combining

Apriori algorithm and constraint-based genetic

algorithm for tree induction for aircraft electronic

ballasts troubleshooting, Lecture Notes in Computer

Science. pp. 381-384.

Chiu, N. H., Huang, S. J., 2007. The adjusted analogy-

based software effort estimation based on similarity

distances, Journal of Systems and Software. Vol. 80,

pp. 628-640.

Milosovski, G., Bil, C., and Kosevski, M., 2008.

Application of expert systems to aircraft accident

investigation, 46th AIAA Aerospace Sciences Meeting

and Exhibit, Reno, Nevada, Jan. pp. 7-10.

O'Hare, D., Chalmers, D., and Scuffham, P., 2003. Case-

control study of risk factors for fatal and non-fatal

injury in crashes of civil aircraft, Aviation, Space, and

Environmental Medicine. vol. 74, pp. 1067-1072.

Paul, R. A., Challagulla, V. U. B., Bastani, F. B., and Yen,

I. L., 1993. A memory-based reasoning approach for

assessing software quality, Computer Software and

Applications Conference.

Taneja, N., Wiegmann, D. A., 2003. Analysis of injuries

among pilots killed in fatal helicopter accidents,

Aviation, Space, and Environmental Medicine. vol.

74, pp. 337-341.

Weng, C. T., Ho, C. S., and Lan, C. E., 2003.

Aerodynamic model estimation and analysis for a jet

transport in a landing accident, AIAA Atmospheric

Flight Mechanics Conference and Exhibit, Austin,

Texas, August.

ICSOFT 2010 - 5th International Conference on Software and Data Technologies

490