https://doi.org/10.20965/jaciii.2017.

p0467

Analysis of Near-Miss Ship Collisions Using Logistic Regression

Paper:

Study on the Analysis of Near-Miss Ship Collisions

Using Logistic Regression
Kwang-Il Kim∗ , Jung Sik Jeong∗ , and Byung-Gil Lee∗∗
∗ InternationalMaritime Transportation Science, Mokpo National Maritime University
91 Haeyangdaehag-ro, Mokpo, Jeon-Nam 58628, Republic of Korea
E-mail: [email protected], [email protected]
∗∗ Electronics and Telecommunications Research Institute

218 Gajeong-ro, Yuseong-gu, Daejeon 34129, Republic of Korea

E-mail: [email protected]
[Received March 20, 2016; accepted January 4, 2017]

Generally, risk assessment for a ship collision can be ing CPA, TCPA, and the encounter angle between vessels.
performed by analyzing the trajectories of two ships This dynamic model only calculates collision risk for en-
as they get close to each other. A near-miss collision counter situations.
between ships is an undesired event that did not re- The second part is made up of the consequences of
sult in collision, but had a high risk of doing so. Due encounter situations, such as collision accidents or near-
to the high frequency of these occurrences, many ac- misses. Collision accident data contains sufficient expla-
tual accident data samples can be obtained. In this nation regarding an encounter situation. Additionally, it
paper, we extract various variables related to near- is categorized according to ship size and type, encounter
miss collisions from this data, such as Distance to Clos- type (i.e. head on, cross, overtake), human factors in-
est Point of Approach (DCPA), Time to Closest Point volved in the collision accident, etc.
of Approach (TCPA) and Collision Avoidance Vari- By looking at previous studies on the usage of ship
ance (CAV). To assess near-miss collision risk, logistic collision accident data as dependent variables, Roeleven
regression analysis is performed by categorizing en- et al. [5] calculated the probability of a collision acci-
counter types based on ship trajectories collected over dent through a casuistic approach by using the binary
4 months in coastal water areas. logit model of Generalized Linear Models (GLM). In this
model, collision accident data contains a two-value de-
pendent variable indicating whether or not a collision ac-
Keywords: ship encounter variables, near-miss ship col- cident occurs. Jin et al. [6] also employed a binary logit
lision, logistic regression model for modeling the probability of fishing ship ac-
cidents. It used wind speed, ship length, navigating or
working regions, seasons, etc. obtained from dichoto-
1. Introduction mous accident data.
Despite the many advantages of collision accident data,
Ship collisions are one of the most common maritime there is still a significant lack of samples for statistical
accidents at sea. In order to reduce collision numbers, analysis. As Debnath [7] and Inoue et al. [8] pointed out,
ship collision risk assessment has become an important it is necessary to have a database containing a sufficiently
issue for maritime safety and traffic management. Many large number of collision accident cases to draw statisti-
researchers have reported that conflicts in navigation in- cally sound inferences from the analysis of collision acci-
tentions account for a substantial portion of ship colli- dent records.
sions [1, 2]. In general, risk assessment for ship collisions Table 1 displays statistical data on the consequences
can be performed by analyzing the trajectories of ships as of encounter situations over a one-year time period in the
they near each other, using Automatic Identification Sys- coastal water areas of Busan Port and Busan New Port
tem (AIS) dynamic information. It is reported that collid- located in the south-eastern part of South Korea. The
ing ships typically exhibit unsafe movements [3]. number of reported collision accident is only three per
A ship encounter situation consists of two main parts. year [9]. Near-misses, however, account for 1,278 cases,
The first part is made up of collision predictors such as and close quarter situations account for 15,114 cases;
Distance to Closest Point of Approach (DCPA) and Time these numbers are sufficient for statistical analysis. A
to Closest Point of Approach (TCPA); these are proximity near-miss collision is an undesired event that did not re-
indicators for the Closest Point of Approach (CPA). They sult in collision, but had a high risk of doing so. A close
represent temporal and spatial margins between two ships quarter situation is a sequence of events and/or conditions
prior to a collision. between ships that had the potential to result in a colli-
J. M. Mou et al. [4] proposed a dynamic risk model us- sion [10]. The encounter distance of a close quarter situa-

Vol.21 No.3, 2017 Journal of Advanced Computational Intelligence 467

and Intelligent Informatics
© Fuji Technology Press Ltd. Creative Commons CC BY-ND: This is an Open Access article distributed under the terms of
the Creative Commons Attribution-NoDerivatives 4.0 International License (http://creativecommons.org/licenses/by-nd/4.0/).
Kim, K.-I., Jeong, J. S., and Lee, B.-G.

Table 1. The number of consequences of encounter situa-

tions in target area.
Number per year
Collision accident 3
Collision near-miss 1,278
Close quarter situation 15,114

tion is typically larger than that of a near-miss.

To overcome the low sample problem, several stud-
ies have examined near-miss ship collisions (called near-
miss) instead of actual collisions [11, 12]. Near-miss en-
counters between ships occur frequently enough to enable
statistical analysis. It is also possible to obtain ship size
and type information, as well as encounter types from the
AIS device, equivalent to collision accident reports con- Fig. 1. Categorization of encounter type.
tents except human errors.
In this paper, we extract statistical variables related to
near-misses risk from the consequences of ship encounter DCPA(t) =
situations. In order to obtain a model equation, ship en- 
counter cases related to near-misses are examined from [(y j−yi )+(yj−yi )×TCPA]2+[(x j−xi )+(xj−xi )×TCPA]2
the analysis of ship trajectories. In this study, because . . . . . . . . . . . . . . . . . . . (2)
near-miss data is presented as a dichotomous variable
with values of “near-miss” and “no near-miss,” logistic re- Another predictor variable in a collision situation is the
gression analysis is appropriate for finding the best model collision avoidance action. A collision avoidance action
describing the relationship between a near-miss event and can lower collision risk by increasing DCPA and TCPA.
the ship collision assessment variables from an encounter Therefore, it is closely correlated with the occurrence of
situation. a near-miss. In this study, we propose the concept of
The remainder of this paper is organized as follows. Collision Avoidance Variance (CAV). It is a dichotomous
Section 2 describes the extraction of ship encounter vari- variable derived by checking if course deviation is over
ables for the regression model. In Section 3, independent 15◦ . CAV(1) indicates less than 15◦ course variance, and
and dependent variables are extracted from AIS ship tra- CAV(2) indicates greater than 15◦ course variance.
jectory data in the target water area and used to derive a The other relevant variable for ship collisions is en-
model equation using logistic regression. Finally, conclu- counter type. An encounter can be classified as head-
sions are drawn in Section 4. on, cross, or overtake, as defined in the Collision Regu-
lations [14]. Using a ship’s heading as a bearing of 0◦ , a
head-on situation is −5◦ to 5◦ , a cross situation is 5◦ to
112.5◦ and −5◦ to −112.5◦, and an overtake situation is
2. Ship Encounter Variables
112.5◦ to 180◦ and −112.5◦ to −180◦ . These encounter
types are illustrated in Fig. 2.
2.1. Predictors of Collision
Generally, when one ship is about to collide with an- 2.2. Near-Miss Variable
other, the navigators make a plan for collision avoidance
by considering the magnitude of DCPA and TCPA to de- A near-miss collision means that ship navigation almost
termine an appropriate collision avoidance action. resulted in a collision or a near-passing course, but colli-
DCPA is the distance between two ships at their CPA. sion was avoided through navigator action or luck [12].
TCPA is the time required to move from the current ship To discriminate near-miss events from other encoun-
position to the CPA. In Fig. 1, DCPA and TCPA are both ters, Kim et al. [12] and Goerlandt et al. [11] employed
illustrated in a cross situation encounter [13]. the ship elliptical domains proposed by Fujii [15] as near-
Let the own ship and the target ship be approaching miss criteria. Fig. 3 shows the near-miss and no near-miss
each other from their current positions of (xi , yi ) and areas between nearby ships. A near-miss event variable is
(x j , y j ) with latitudinal and longitudinal components of dichotomous with values of occurrence (value = 1) and
(xi , yi ) and (xj , yj ) according to each ship’s course and no-occurrence (value = 0) determined using this ship do-
velocity. main criteria.
In this encounter situation, DCPA and TCPA can be When selecting near-miss events from other encoun-
derived as follows: ters, analysis of ship trajectories is performed whether or
not a target ship’s position falls inside the owned ship’s
−[(y j − yi )(yj − yi ) + (x j − xi )(xj − xi )] domain. Based on ship length (L), an elliptical domain
TCPA(t) = (1)
(yj − yi )2 + (xj − xi )2 in sea speed is drawn with size 8 L in the direction of

468 Journal of Advanced Computational Intelligence Vol.21 No.3, 2017

and Intelligent Informatics
 Analysis of Near-Miss Ship Collisions Using Logistic Regression

Fig. 2. Categorization of encounter types.

Fig. 5. Ship trajectories and close quarter situation points in
the target area.

lated based on the ship’s course (ψ ) using Eq. (4).


|90◦ − ψ | (ψ ≥ 180◦ )
θ= . . . . . . (4)
|270◦ − ψ | (ψ ≤ 180◦ )
After applying Eqs. (3) and (4) to the elliptical domain,
Eq. (5) is used to discriminate near-miss events between
the own and target ships. If the value on the left is less
Fig. 3. Concept of a near-miss event illustrated by an ellip- than 1, a near-miss event occurred, otherwise, no near-
tical domain. miss occurred.
(cos θ × (xc − xo ) + sin θ × (yc − yo ))2
+
(4 × Lo )2
(sin θ × (xc − xo ) − cos θ × (yc − yo ))2
≤1 . (5)
(1.6 × Lo )2

3. Methodology

3.1. Target Water Area

The target water area for this study is the coastal water
area around Busan port and Busan New port located in the
south-eastern part of South Korea. We use ship trajectory
data from a 4 month time period. This coastal water area
is suitable for assessing ship collision risk because there
Fig. 4. Discrimination of the ship elliptical domain. are no obstructions or route regulations. Therefore, near-
miss situations occur more frequently compared to port
water areas. Fig. 5 shows ship trajectories and close quar-
ter situation points (CPA < 0.5 nautical miles and TCPA
the ship’s course, and 3.2 L perpendicular to the ship’s < 5 minutes) in the target area.
course [15]. Fig. 4 represents discrimination of near-miss
events by rotating the own ship’s elliptical domain and
comparing the target ship’s position to the boundary. 3.2. Extraction of Model Variables
Because the elliptical domain depends on the ship’s Ship trajectory data received from an AIS device con-
course, it must be rotated to match the course direction. sists of dynamic and static information. The dynamic in-
Using Eq. (3), a coordinate (x, y) on the ship domain formation includes vessel movement information such as
boundary can be rotated to (x , y ). ship position, course, and speed. Encounter information
x = cos θ × x + sin θ × y between navigating ships is not included [16]. Therefore,
we must extract encounter variables from ship trajectory
y = sin θ × x − cos θ × y . . . . . . . . (3)
data. Fig. 6 illustrates the process of extracting ship en-
Where, angle of intersection with the x-axis (θ ) is calcu- counter variables.

Vol.21 No.3, 2017 Journal of Advanced Computational Intelligence 469

and Intelligent Informatics
Kim, K.-I., Jeong, J. S., and Lee, B.-G.

Table 2. Description of near-miss variables.

Variable Variable
Description p-value
Name Type
DCPA/L Mean DCPA /
0.000
(X1 ) Own Ship length
TCPA Independent
TCPA at 0.5 nm 0.001
(X2 ) Variables
CAV(1): under 15◦
CAV
CAV(2): more than 0.000
(X3 )
15◦
1: Head-on
Encounter Selection
2: Cross –
Type Variable
2: 3: Overtake
Near-miss Dependent 0: No Near-miss
–
Event Variable 1: Near-miss

Table 3. The resulting coefficients (Odds Ratio) from logis-

tic regression.
DCPA/L TCPA CAV Constant
(β1 ) (β2 ) (β3 ) (α )
−2.007 −0.492 −1.135 3.7
Head on
(0.134) (0.611) (0.321) (40.45)
−1.414 −0.075 −1.055 2.125
Cross
(0.243) (0.928) (0.348) (8.37)
−0.905 −0.256 −1.776 3.237
Overtake
(0.404) (0.774) (0.169) (25.45)
( ) : Odds Ratio

is DCPA based on own ship length, and this is used be-

cause domain size depends on ship length. TCPA is cal-
culated when the distance between ships reaches 0.5 nm
because that is the close quarter threshold. CAV is deter-
mined by the difference between maximum course change
and minimum course change in a close quarter situa-
Fig. 6. Ship encounter data extraction module. tion [18].
Although the encounter type variable is not statistically
significant to a near-miss event, it does has an effect on
DCPA/L, TCPA, and CAV variables (see Table 3). It is
First, ship trajectory data is arranged in chronolog- used as a selection variable for separating encounters into
ical order, and then data from adjacent ships is com- three groups (head-on, cross, overtake).
bined to form a ship encounter at the same time i. Data
is combined based the close quarter situation criteria of
DCPA being less than 0.5 nautical miles, and TCPA be- 3.3. Applying to the Logistic Regression
ing less than 10 minutes [9]. Second, DCPA/L, TCPA, Logistic regression is a type of linear regression used to
CAV, and encounter type are calculated. Position of an analyze dichotomous dependent variables, such as near-
encounter are determined using Dead Reckoning Posi- miss occurrence. It is based on the construction of a sta-
tions [17]. Third, the module determines the minimum tistical model that describes the relationship between ob-
distance in an encounter and decides whether or not a served dependent and independent variables, also called
near-miss event occurred using Eq. (5). Finally, ship en- response and explanatory variables [19]. It can not only
counter and near-miss event variables are saved to the be used to predict posteriori probability, but also for clas-
database. After searching through all ship trajectories, re- sification of data samples. Therefore, it can be used to
gression coefficients are calculated using the Maximum calculate posteriori probability using categorical and con-
Likelihood method. tinuous variables through a logistic function.
Table 2 displays the descriptions of near-miss variables In this study, binomial logistic regression is applied to
and their statistical significance (p-value) in this analysis. test for an association between the dependent variable of a
DCPA/L, TCPA, and CAV are used as dependent vari- near-miss and the independent variables of related data in
ables. These variables satisfy the statistical significance an encounter. By applying the logistic function in Eq. (6),
threshold (p-value < 0.05) for a near-miss event. DCPA/L near-miss probability becomes a value between 0 and 1,

470 Journal of Advanced Computational Intelligence Vol.21 No.3, 2017

and Intelligent Informatics
 Analysis of Near-Miss Ship Collisions Using Logistic Regression

regardless of the size of the independent variables.

e(α +β1 X1 +···+βn Xn )
NP = . . . . . . . (6)
1 + e(α +β1 X1 +···+βn Xn )
For estimation of the model coefficient value of Xi , we use
a logit transformation function similar to a linear regres-
sion model, shown in Eq. (7):
 
NP
log it(NP) = ln = α + β1 X1 + · · · + βn Xn (7)
1 − NP
The regression equation coefficients (β ) are estimated
using the Maximum Likelihood method proposed by Fig. 7. Near-miss probability curves in head-on situation.
Cox [20]. The primary output from the logistic regression
model is an Odds Ratio (OR), or probability, of a near-
miss event. The OR is the relative amount by which the
odds of an outcome increase (OR greater than 1.0) or de-
crease (OR less than 1.0) when the value of the predictor
is increased by 1 unit.
Using the independent and dependent variables from
ship encounter data, logistic analysis is performed by the
SPSS package. Table 3 displays coefficients (β ), and OR.
In order to calculate near-miss probability, regression
model equations are derived using the regression coeffi-
cients as shown in Eqs. (8)–(10).
e(3.7−2.01X1 −0.49X2 −1.14X3 )
NPhead−on = . . (8)
1 + e(3.7−2.01X1 −0.49X2 −1.14X3 ) Fig. 8. Near-miss probability curves in cross situation.

e(2.13−1.41X1 −0.08X2 −1.06X3 )

NPcross = . . . (9)
1 + e(2.13−1.41X1 −0.08X2 −1.06X3 )
e(3.24−0.91X1 −0.26X2 −1.78X3 )
NPovertake = . (10)
1 + e(3.24−0.91X1 −0.26X2 −1.78X3 )
Where, X1 : DCPA/L, X2 : TCPA, X3 : CAV (1) =
0,CAV (2) = 1.

3.4. Discussion
The DCPA/L independent variable has a negative effect
on near-miss probability in all encounter situations (OR in
head-on: −2.007 times, OR in cross: −1.414 times, OR Fig. 9. Near-miss probability curves in overtake situation.
in over-take: −0.905 times) when it is decreased by 1.
The TCPA independent variable also has a negative effect
on near-miss probability in all encounter situations (OR
in head-on: −0.492 times, OR in cross: −0.075 times, CAV(1) and CAV(2) for near-miss probability is signifi-
OR in over-take: −0.256 times) when it is decreased by cantly higher for overtake situations compared to the other
1. For the CAV independent variable, CAV(1) has a nega- encounter types.
tive effect when compared with CAV(2) for all encounter
situations (OR in head-on: −1.135 times, OR in cross:
−1.055 times, OR in over-take: −1.776 times). 4. Conclusion
Figures 7–9 shows near-miss probability distribution
curves calculated using the regression coefficients values We utilized ship encounter data including DCPA/L,
for each encounter type. The X-axis is DCPAL, the Y -axis CAV, and encounter type from near-miss events for logis-
is TCPA, and the Z-axis is Near-miss Probability. The tic regression to perform statistical assessment of water-
slopes of the curves are determined by |β |, which is asso- way risk. Near-misses, which are as a dependent variable,
ciated with independent variable sensibility. are substituted for actual accidents due to their relatively
In the distributions of near-miss probabilities, the vari- high occurrence in the available dataset.
ance in head-on and cross situation are higher than in The model coefficients, OR, are obtained by applying
overtake situations. Additionally the difference between regression to the encounter data. The resulting model

Vol.21 No.3, 2017 Journal of Advanced Computational Intelligence 471

and Intelligent Informatics
Kim, K.-I., Jeong, J. S., and Lee, B.-G.

equations are useful for analyzing risk factors posed by [19] D. R. Collet, “Modelling Binary Data,” Chapman and Hall, 1991.
independent variables, and calculating near-miss proba- [20] D. K. Cox and E. J. Snell, “The Analysis of Binary Data,” London,
1970.
bility.
The results will support vessel traffic service operators
or ship navigators in collision avoidance and real-time
traffic control. Additional encounter data and the addition Name:
Kwang-Il Kim
of other risk variables will improve model classification
accuracy in the future. Affiliation:
Computer Science, Chungbuk National Univer-
sity
Acknowledgements
This research was supported by Basic Science Research Program
through the National Research Foundation of Korea (NRF) funded
by the Ministry of Education (NRF-2016R1A6A3A 11935806). Address:
This work was supported by ETRI through the Maritime Safety & 1 Chungdae-ro, Seowon-gu, Cheongju, Chungbuk 28644, Republic of
Maritime Traffic Management R&D Program of the MPSS (Min- Korea
istry of Public Safety and Security) and KIMST (2009403, Devel- Brief Biographical History:
opment of next generation VTS for maritime safety). 2005 Received the Bachelor of Science in Department of Maritime
Transportation System from Mokpo National Maritime University
2005-2009 Navigator-Deck Officer, Ssangyong Shipping, Co. Ltd.
2010- VTS Officer, Yeosu VTS, Ministry of Public Safety and Security,
References: Republic of Korea
[1] J. S. Jeong, G. K. Park, and K. I. Kim, “Risk Assessment Model 2012 Received the Master degree in Department of Maritime
of Maritime Traffic in Time-Variant CPA Environments in Water- Transportation System from Mokpo National Maritime University
way,” J. Adv. Comput. Intell. Intell. Inform. (JACIII), Vol.16, No.7, 2012 Received the Doctor degree in Department of Maritime
pp. 866-873, 2012.
Transportation System from Mokpo National Maritime University
[2] R. M. Darbra and J. Casal, “Historical analysis of accidents in sea-
ports,” Safety Science, Vol.42, No.2, pp. 85-98, 2004. Main Works:
[3] T. Hong, “Development of a System for Transmitting a Navigator’s • “Risk Assessment Model of Maritime Traffic in Time-Variant CPA
Intention for Safe Navigation,” Int. J. of Fuzzy Logic and Intelligent Environments in Waterway,” J. Adv. Comput. Intell. Intell. Inform.
Systems, Vol.14, No.2, pp. 130-135, 2014. (JACIII), Vol.16, No.7, pp. 866-873, 2012.
[4] J. M. Mou, C. van der Tak, and H. Ligteringen, “Study on collision Membership in Academic Societies:
avoidance in busy waterways by using AIS data,” Ocean Engineer- • Korea Institute of Intelligent Systems
ing, Vol.37, pp. 483-490, 2010. • Korean Institute of Navigation and Port Research
[5] D. Roeleven, M. Kok, H. L. Stipdonk, and W. A. de Vries, “In-
land Waterway Transport; Modeling the Probability of Accidents,”
Safety Science, Vol.19, pp. 191-202, 1995.
[6] D. Jin, H. L. Kite-Powell, E. Thunberg, A. R. Solow, and W. K. Tal-
ley, “A Model of Fishing Vessel Accident Probability,” J. of Safety
Research, Vol.33, No.4, pp. 497-510. 2002. Name:
[7] A. K. Debnath, “Traffic Conflict based Modeling of Collision Risk Jung Sik Jeong
in Port Waters,” Ph.D. dissertation, National University of Singa-
pore, Singapore, pp. 29-30, 2009.
[8] K. Inoue and M. Kawase, “Innovative Probabilistic Prediction of Affiliation:
Accident Occurrence,” Int. J. on Marine Navigation and Safety of Mokpo National Maritime University
Sea Transportation, Vol.1, No.1, 2007.
[9] Korea Maritime Safety Tribunal, Maritime Accidents Statistics
2011-2016, http://www.kmst.go.kr [accessed Mar. 1, 2016], 2017.
[10] C. Jean-Charles, “Close quarter situations reporting by VTS,”
Rights and Safety of Marime and Ocean Activities Speciality,
pp. 38-39, 2004.
[11] F. Goerlandt, et al., “Analysis of the Near-collisions using AIS data Address:
for the selected locations in the Baltic Sea,” EfficienSea Tech. Rep., 91 Haeyangdaehag-Ro, Mokpo City, Jeon-Nam, 530-729, Republic of
2012. Korea
[12] K. I. Kim, J. S. Jeong, and G. K. Park, “A Study on Near-miss Inci- Brief Biographical History:
dents from Maritime Traffic Flow by Clustering Vessel Positions,” 1987 Received the Bachelor of Science in Department of Nautical
J. of Korean Institute of Intelligent Systems, Vol.24, No.6, pp. 603-
608, 2014. Science from Korea Maritime University
[13] A. K. Debnath and H. C. Chim, “National Traffic Conflict Tech- 1993 Received the Master Degree in Department of Electronic and
nique : A Proactive Approach to Quantitative Measurement of Col- Communication from Korea Maritime University
lision Risks in Port Waters,” The J. of Navigation, Vol.63, pp. 137- 1993-1996 Senior Researcher, Korea Telecom R&D Center
152, 2010. 2001 Received the Doctoral Degree in the Department of Electrical and
[14] IMO, COLREG: Convention on the international regulations for Electronic Engineering from Tokyo Institute of Technology
preventing collisions at sea 1972, The International Maritime Or- 2001-2002 Senior Researcher, Korea Telecom R&D Center
ganization, 2002. 2002 Associate Professor, Faculty of Maritime Transportation
[15] Y. Fujii and K. Tanaka, “Traffic Capacity,” The J. of Navigation, Main Works:
Vol.24, pp. 543-552, 1971.
• Parameter estimation, statistical risk assessment, maritime traffic
[16] E. K. Kim, J. S. Jeong, G. K. Park, and N. K. Im, “Characteristics of
Ship Movements in a Fairway,” Int. J. of Fuzzy Logic and Intelligent engineering, collision avoidance
Systems, Vol.12, No.4, pp. 285-289, 2012. Membership in Academic Societies:
[17] J. S. Kim, J. S. Jeong, and G. K. Park, “Utilization of Planned • Korea Institute of Intelligent Systems
Routes and Dead Reckoning Positions to Improve Situation Aware- • Korean Institute of Navigation and Port Research
ness at Sea,” Int. J. of Fuzzy Logic and Intelligent Systems, Vol.14, • The Institute of Electrical and Electronics Engineers (IEEE) Signal
No.4, pp. 288-294, 2014. Processing Society
[18] K. I. Kim, J. S. Jeong, and G. K. Park, “A Study on the Analysis
of Ship Encounter Data using Logistic Regression,” The 16th Int.
Symp. on Advanced Intelligent Systems, pp. 235-242, 2015.

472 Journal of Advanced Computational Intelligence Vol.21 No.3, 2017

and Intelligent Informatics
 Analysis of Near-Miss Ship Collisions Using Logistic Regression

Name:
Byung-Gil Lee

Affiliation:
Electronics and Telecommunications Research
Institute (ETRI)

Address:
218 Gajeong-ro, Yuseong-gu, Daejeon, 34129, Republic of Korea
Brief Biographical History:
1991-2001 Senior Research Engineer, Dacom R&D Center
2001-present Principle Member of Engineering Staff in Hyper-connected
Communication research laboratory of ETRI
2003 Received the Doctoral Degree in the Department of Electrical and
Electronic Engineering from the Kyungpook National University
Main Works:
• “An intelligent real-time multi-vessel collision risk assessment system
from VTS view point based on fuzzy inference system,” Elsevier Expert
Systems with Applications, Vol.40, Issue 4, pp. 1220-1230, 2013.
Membership in Academic Societies:
• Korea Institute of Information and Communication Engineering
• Korean Institute of Navigation and Port Research
• Korean Institute of Communications and Information Sciences
• Korea Multimedia Society

Vol.21 No.3, 2017 Journal of Advanced Computational Intelligence 473

and Intelligent Informatics

Powered by TCPDF (www.tcpdf.org)