Identification of crash-prone segment on Cipularang toll road

Traffic safety is an important issue in the field of transportation. Toll roads are one of the infrastructures that are prone to crashes, because on toll roads drivers can drive their vehicles at high speeds so there is a high risk of having an crash, as happens every year on Cipularang Toll Road. This research will determine the crash-prone segments on Cipularang Toll Road section using the network screening method. Crash and traffic data for the last five years (2018-2022) were obtained from PT Jasa Marga Purbaleunyi. Identification of crash-prone segments carried out using ten performance indicators contained in HSM (2010), namely average crash frequency, crash rate, critical rate, Equivalent Property Damage Only average crash frequency, relative severity index, excess predicted average crash frequency using method of moment, Level of Service of Safety, excess predicted average crash frequency using SPF, probability of specific crash type exceeding threshold proportion, and expected average crash frequency with Empirical Bayes adjustment. The analysis results show that the five segments with the highest priority for treatment are KM 91-92, KM 92-93, and KM 86-87 (Jakarta-bound direction), KM 120-121 and KM 101-102 (Bandung-bound direction).


Introduction
Land transportation is one of the solutions to fulfill human needs by moving goods or services with the infrastructure used, namely roads.Road infrastructure is usually prone to crashes, one of which is the freeway or toll road.With the "freeway" status, it does not guarantee that traffic crash problems can be overcome.Crash mostly happens because freeways spur drivers to use high speeds to exceed the permitted limits.One of the factors that cause crashes is road factors (geometric) and the lack of road safety furnitures.It is necessary to do road safety management [1].
One of the toll roads that are prone to crashes is the Cipularang Toll Road (Cikampek -Purwakarta -Padalarang).Crashes on this toll road occur every year, causing the formation of several crash-prone points on this toll road.On this toll road, front and rear crash crashes generally occur.This resulted in a high number of victims who indicated serious or minor injuries to death.
Several studies on the ranking of crash-prone segments on the Cipularang Toll Road section have been carried out using various methods.In this study, crash-prone segments will be identified with the data from 2018 to 2022 using the network screening method, which refers to the Highway Safety Manuals (2010).
The network screening method will obtain comprehensive results because the lowest number of crashes in a segment can be adequately identified due to RTM bias considerations.So that the ranking results will show whether there has been a change in the crash-prone segment from previous studies, which indicates effectiveness in handling crashes on the Cipularang Toll Road section before.

Literature Review
Several studies about the analysis of crashes on the Cipularang Toll Road have already been carried out.The studies also analyzing crash prediction modeling and crash cost.

Crash Prone Segment Identification
There are several studies regarding the identification of crash-prone segments on the Cipularang Toll Road using a variety of different methods.In 2017, Damanhuri [3] research started by identifying vehicle travel behavior, blackspot location geometry, and the influence of travel behavior on blackspot location geometry.Then in 2019 Hanafi et al. [4] analyzing the crash-prone segment by using Korlantas weighting method and Deni with Mayani [5] analyzing the crash-prone segment using a descriptive method.Meanwhile, in the research of Andre Jonathan et al. [6] in 2020, a one-way Annova comparative test was carried out by including the crash rate calculation in finding crash-prone locations.In general, the results of previous research indicated that the crash-prone segment on the Cipularang Toll Road section is located at KM 91 to KM 93 Jakarta-bound direction.

Crash Prediction Model
Research conducted by Kusumawati and Rakhmat (2011) [7] used the Generalized Poisson (GP) regression model which is generalized to the standard Poisson regression model.The performance of the model will be compared with the model developed using the Negative Binomial (NB) regression model.Parameters used in GP crash prediction modeling and NB regression include average daily traffic, the number of horizontal curves, and the existence of a median with certain specifications.Here are the basic equations used for modeling.
Where, μ = Expected crashes (incidents/ Where, μ = Expected crashes (incidents/3 years) LHRT = Annual Average Daily Traffic (3 years average in vehicles/day) D = Degrees of curvature (radians) MED = Existence of a median with a width < 2.5 m and a height < 1.75 m (1 if the median meets these specifications, 0 otherwise) Then in the research of Kusumawati et al. (2013) [9] an crash prediction model was developed for case studies of the Palimanan -Kanci, Tangerang -Merak, and Purbaleunyi toll roads.This study aims to develop a general crash prediction model for toll roads in Indonesia by considering the relationship between crash frequency and traffic flow, as well as various road and environmental geometric characteristics.This model can also be used to identify blackspot locations on toll roads with crash data from 2007 to 2010.The following is the general equation resulting from this study.

Crash Cost
In Meydita and Kusumawati [10] research on 2013, an analysis of the crashes cost that occurred on the toll road was carried out using the human capital method approach.This approach will calculate the crash costs based on direct and indirect costs.The classification of crash costs is based on the number of vehicles involved and the level of crash fatality, namely fatal, serious injury, light injury, or material loss.
The crash costs in this study are higher because what is calculated is the unit crash cost for toll roads while in Pd T-02-2005-B the unit crash costs for urban and inter-urban roads which can involve motorcycles, the losses are materially cheaper.Apart from that, this research also takes into account the costs of delays or congestion so that it is more comprehensive compared to Pd T-02-2005-B.The following is a recapitulation of toll road crash costs resulting from this study.

Methodology
The data processing uses the network screening method based on the American Association of State Highways and Transportation (2010).The data to be processed includes crash and traffic data obtained from PT Jasa Marga Purbaleunyi in the period of 2018 to 2022.The network screening method will review a transportation network that is intended to identify and rank the most likely sites to realize a reduction in the frequency of crashes by implementing preventive measures.
The network screening method consists of five processes which can be explained as follows.a. Determining the focus of this analysis, which is to identify and rank sites that have the potential for improvement in reducing the number of crashes.b.Identifying Networks and Establishing Reference Populations.In this study, the reference population was determined from the start which is the Cipularang Toll Road section from KM 67 to KM 121.6, which passes through the Cikampek, Purwakarta, and Padalarang areas.c.Determining performance measures that are selected based on the availability of existing data.
Based on HSM (2010), all performance measures require crash data and road information for categorization.Things that make it different include the need for traffic data on several performance measures, including crash rates, critical rates, level of service of safety, and others.
In addition, there are several performance measures that require calibrated safety performance measures and overdispersion parameter data, such as excess predicted average crash frequency using Safety Performance Functions (SPFs) and Equivalent Property Damage Only (EPDO) average crash frequency with EB adjustment.Other considerations include the EPDO weighting factor and relative severity indices contained in several performance measures.d.Determining a Screening Method.In this analysis, the simple ranking method is used to rank 1km crash-prone segments on Cipularang Toll Road Bandung-bound direction (A-direction) and Jakarta-bound direction (B-direction).e. Screen and Evaluation of Results.The segment most likely to reduce the frequency of crashes is the most suitable for the selected performance measures.In this study, the selection of crashprone segment will be limited to five segments throughout the Cipularang Toll Road.

Results
The steps taken before determining crash-prone segments are determining performance measures.In determining performance measures there are some criteria for selecting them.One of them is the availability of data and input.The required data is available, including crash and traffic volume data.In addition, a safety performance function (SPF) and crash cost data are needed, which is one of the criteria for data availability.The SPF equation used is contained in equation ( 2) while the crash cost data is shown in Table 1.
Based on these considerations, there are ten performance measures that can be used in this study.The following are the performance measures used to rank crash-prone segments on the Cipularang Toll Road section.For a more detailed calculation method, see the Highway Safety Manual (2010) [2].

Average Crash Frequency
In a ranking based on the average crash frequency, the segment that has the largest total number of crashes in the 5-year range, from 2018 to 2022, will be given the highest ranking.The following is the ranking of ten most hazardous segments based on the average crash frequency.

Crash Rate
For performance measure crash rate, crash frequency will be measured by traffic volume.The exposure used for sections or segments is road vehicles or VMT (Vehicle-Miles Travelled) over a fiveyear period.The following is the calculation and ranking results based on the crash rate in each segment, which is limited to ten crash-prone segments.

Critical Rate
The ranking is based on the crash rate at each location compared to the critical crash rate using the methods from the Highway Safety Manual.The following is the result of the calculation and ranking based on the difference in the crash rate and critical crash rate in each segment.Weighting factor is assigned to crashes based on severity, namely fatal, serious injury, minor injury, or property damage only (PDO).The weighting factor is calculated relative to the cost of the crash resulting in loss of property or PDO.Crash costs based on severity will produce an EPDO value.Later the segments will be ranked from high to low EPDO scores.Sites at the highest rank can be selected for further investigation.So that the following is the result of calculating the EPDO value and its ranking for all segments on both lines.

Relative Severity Index
In determining the Relative Severity Index or RSI it is necessary to determine the cost of an crash for each level of fatality and the total cost of all crashes for each location.Then the average crash cost per segment is compared with the overall average crash cost.The resulting RSI value will show segments that have more than the average or less than the average.The following is the result of processing the RSI value in each segment which is ranked based on the average RSI value per segment.Overall, the average RSI value is IDR 1,270,569,363.00.So that there are 40 segments that have an average RSI value per segment higher than the average RSI.The average RSI value per segment will be affected by the total number of crashes in that segment and the number of crashes with the highest costs.

Excess Predicted Average Crash Frequency Using Method of Moment
The average crash frequency in each segment adjusted for the variance in crash data and the average crash frequency for the reference population.The observed and adjusted average crash frequency for that location is compared to the average crash frequency for the entire segment.The following is a recapitulation of the calculation of the PI value for each segment per kilometer.In the data processing, a comparison will be made between the average crash frequency observed with the average crash frequency from the SPF.The difference between the observed and predicted crash frequencies will be the excess of the crash frequencies estimated using the SPF.If the excess frequency is more than zero, the segment will have the highest ranking based on the excess value.The following is the result of calculating and ranking the segments based on the excess value.The predicted average crash frequency for segments with similar characteristics is predicted from the SPF calibrated to local conditions.The observed crash frequency was compared with prediction of average crash frequency.Each location is placed into one of the four LOSS classifications on HSM (2010) [2].The following is the result of calculating the LOSS value for all segments per kilometer.

. Probability of Specific Crash Types Exceeding Threshold Proportion
A threshold proportion (pi*) will be selected for each population based on the proportion of types of crash or severity.Then the variance value will be calculated for the reference to calculate the alpha and beta parameters used to find the target crash probability.This probability value will be the basis for the ranking of the segment.On the probability calculation, data of fatal and injury crashes will be used.The ranking based on the probability shown in the following table.It will be weighted from the observed average crash frequency and the expected crash frequency from SPF with the EB method.The following is the result of the calculation and ranking process based on the estimated average crash frequency adjusted for EB in the last year, which is 2022.

Selection of Top 5 (Five) Segments Ranking Results
The following is a recapitulation of the ranking results from all performance measures, which are limited to 10 segments.PM on the table refers to Performance Measures and the number refers to the order in the previous section.Considering all existing performance measures and the frequency of ranking of each performance measure, here are the five most crash-prone segments on the Cipularang Toll Road section, using crash data and traffic data from 2018 to 2022.

Discussions
Segment ranking is based on the frequency of fulfilling each performance measure and being in the top five positions.As previously explained, the segment higher up on the list will be considered as the segment which will most likely to benefit from preventive measures intended to reduce the frequency of crashes or severities.
Ranking must comply with existing performance measures by prioritizing the top five rankings in each ranking result.B91-92 segment occupies the first position because out of ten performance measures, this segment occupies the first position in eight performance measures.In the second rank is the B92-93 segment, which has the second highest rank for all performance measures.
In third and fourth place, segments A120-121 and A101-102 are in the same position.However, the A120-121 segment is ranked higher because the level of compliance with performance measures and its position in the top five rankings is higher than the A101-102 segment, as in the EPDO Average Crash Frequency and Excess Predicted Average Crash Frequency Using SPFs.Based on these considerations, the A120-121 segment received a higher ranking because it could benefit from the preventive measures provided.
The results of the ranking that have been carried out are in accordance with previous research.This shows that the location of crash-prone segments on the Cipularang Toll Road has stayed the same.It is also indicated that the treatment provided before has not been successful in dealing with crashes on the Cipularang Toll Road section.

Conclusions
Based on the results of ranking crash-prone segments using the network screening method, the most crash-prone segments on the Cipularang Toll Road section based on data from 2018 to 2022 are KM 91 -92 Jakarta-bound direction, KM 92 -93 Jakarta-bound direction, KM 120 -121 Bandung-bound direction, KM 101 -102 Bandung-bound direction, and KM 86 -87 Jakarta-bound direction.In terms of previous studies, there has been an increase in the number of crashes and the number of crash victims.But the crash-prone locations are still between KM 91 to KM 93.This shows previous treatments applied to the locations were not able to improve the road safety condition that the locations remain to be the most crash-prone locations.The use of the network screening method helps in determining crash-prone segments in a more comprehensive manner.Because with RTM bias considerations, all segments with lowest frequency can be identified properly.In addition, the existence of various performance measures used in the rating process helps to review the selection of crash-prone segments from several exposures used.

Recommendations
Suggestions that can be given for further research regarding the ranking of crash-prone segments is to carry out an analysis using the relationship between the number of crashes and the types of vehicles involved.Research can be continued by determining alternative recommendations for crash-prone segment.
crashes (incidents/3 years) AADT = Annual Average Daily Traffic (3 years average in vehicles/day) NTIK = Number of horizontal curves W = Track width (m)

Table 1 .
Indonesian Toll Road Crash Costs.

Table 2 .
Average Crash Frequency Data Ranking Results.

Table 3 .
Crash Rate Data Ranking Results.

Table 4 .
Critical Rate Data Ranking Results.

Table 5 .
EPDO Average Crash Frequency Data Ranking Results.

Table 6 .
RSI Data Ranking Results.

Table 7 .
PI Data Ranking Results.

Table 8 .
Excess Predicted Average Crash Frequency Using SPFs Data Ranking Results.

Table 9 .
LOSS Data Ranking Results.

Table 10 .
Target Crash Probability Data Ranking Results.

Table 11 .
Expected Average Crash Frequency with EB Adjustment Data Ranking Results.