Journal of the Civil Engineering Forum.
May 2026, 12.
:189-200 DOI 10.
22146/jcef.
Available Online at https://jurnal.
id/v3/jcef/issue/archive Pedestrian Crossing Safety Model for Unsignalized Three-Leg Intersection Based on User Perception Data Wildan Reza Pahelvi.
Aine Kusumawati.
Taufiq Suryo Nugroho* Faculty of Civil and Environmental Engineering.
Institut Teknologi Bandung.
Bandung.
INDONESIA *Corresponding author: taufiq.
nugroho@itb.
SUBMITTED 14 May 2025 REVISED 04 September 2025 ACCEPTED 04 September 2025 ABSTRACT Recent statistics show an upward trend in road crashes in Indonesia, with pedestrians identified as the most vulner- able group of road users, thus addressing this issue requires evidence-based tools to support decision-making for pedestrian safety improvement.
This study develops a perception-based Pedestrian Intersection Safety Index (PedISI) model using multiple linear regression to estimate safety levels at three-leg unsignalized intersections based on traffic and geometric characteristics.
Unlike previous studies that rely on crash or behavioral data, this research employs user perception data, offering a lower-risk and more flexible means of capturing pedestriansAo subjective evaluations of safety.
The study was conducted at 15 unsignalized three-legged intersections comprising 42 observation points in Cimahi City.
West Java.
Indonesia.
Data were collected on traffic volume, 85th percentile vehicle speed, lane width, and median width, alongside respondentsAo safety ratings derived from on-site video-based surveys.
The results indicate that traffic volume, 85th percentile speed, lane width, and median width significantly influence pedestrian perceptions of crossing safety.
Application of the developed regression model shows that the average perception-based pedestrian safety index at these intersections is 2.
Sensitivity analysis further reveals that reductions in vehicle speed yield the greatest improvements in perceived safety, suggesting that speed management should be prioritized in pedestrian safety interventions.
While geometric factors such as lane and median width also play a role, these must be optimized within design standards to balance safety and traffic performance.
The study highlights the potential of perception-based modeling as a complementary approach for pedestrian safety assessment in data-limited urban environments and provides a framework for future applications incorporating diverse environmental and behavioral contexts.
KEYWORDS Pedestrian Safety.
Perception-based Model.
Unsignalized Intersection.
PedISI Model.
Multiple Linear Regression.
A The Author.
This article is distributed under a Creative Commons Attribution-ShareAlike 4.
0 International license.
1 INTRODUCTION
Indonesia has experienced a dramatic increase in the number of motor vehicles, reaching approximately 153 million units in 2023 (Gaikindo, 2.
This rapid growth has had a considerable impact on road safety, as reflected in the rising number of traffic crashes nationwide.
Traffic crashes not only pose significant public health risks but also result in major economic losses for both the government and the broader community.
In light of these concerns, there is a compelling need to conduct research focused on pedestrian crossing safety as a foundation for implementing more effective safety measures.
Among the least-studied facilities for pedestrian safety are unsignalized three-leg intersections.
Previous studies on pedestrian safety have concentrated more on four-leg intersections (Hashemi et al.
, 2.
, mid-block / median crossings (Avinash et al.
, 2019.
Fitzpatrick et al.
, 2024.
Zhang et al.
, and roundabouts (Xu et al.
, 2.
Yet, threeleg intersections have several characteristics that may impose greater risks to pedestrians when crossing, such as higher turning movement, visibility and gap acceptance, and priority ambiguity .
a Costa et al.
, 2.
This paper addresses this issue by developing a pedestrian crossing safety model specifically for unsignalized three-leg intersections and explicitly considering their characteristics, such as approaching traffic speed.
This paper aims to develop a pedestrian crossing safety model for unsignalized three-leg intersections using a multiple linear regression model that refers to the PedISI (Pedestrian Intersection Safety Inde.
framework developed by Center .
and then later adopted by the Federal Highway Administration (FHWA).
The primary outcome of this research is a safety index that can be used to prioritize interventions at intersections and guide the development of effective strategies for enhancing pedestrian safety in urban environments.
The dependent variable in the model is the average safety rating provided by survey respondents, while the independent variables include the physical and traffic characteristics of the intersections.
One primary data requirement for the PedISI model is behavioral data, such as near-miss incident Journal of the Civil Engineering Forum or crash incident recorded at the study location (Avinash et al.
, 2019.
Fitzpatrick et al.
, 2.
However, this study specifically utilizes user perception data due to the lack of recorded data and the risk associated with collecting behavioral data, such as near-miss incident in the study location (Ogwude et al.
, 2.
On the other hand, the usefulness of using user-perception data is known form literature.
Among its advantages are the low-risk data collection process, the ability to identify hidden safety issues and addressing the problem of the lack of recorded behavioral data (Ihssian and Ismail, 2023.
Kwon et al.
, 2.
Nevertheless, user perception data have their own issues, such as subjectivity bias, limited predictive power, and difficulty in making comparison across sites, particularly if the study is conducted a specific location (Zhu et al.
, 2.
This study uses the Cimahi City and its pedestrians as a case study, which raises questions about the generalizability and transferability of the resulting model to other urban contexts.
The paper addresses these limitations by discussing how case-specific factors may influence the modelAos applicability elsewhere and proposes strategies to enhance generalizability, such as systematic data collection, validation through cross-validation techniques, and the use of objective, widely observable road and traffic
2 LITERATURE REVIEW
Research on pedestrian safety at intersections has been conducted extensively over the years to identify influencing factors and develop effective evaluation models based on regression models.
Early approaches to pedestrian safety primarily focused on analyzing crash trends through police reports and applying statistical measures to improve safety outcomes.
For example.
Jaskiewicz .
introduced a nine-metric evaluation system encompassing aesthetics, safety, and mobility to assess pedestrian environments in Winter Park.
Florida.
Using a fivepoint Likert scale, the study derived overall levels of service (LOS) scores to identify pedestrian deficiencies and suggest physical and policy-based improvements.
A significant contribution in measuring pedestrian safety using a regression model approach was the development of the Pedestrian Intersection Safety Index (PedISI).
This model is an advancement of a macro-level pedestrian safety index developed by Center .
This model was formulated by the Federal Highway Administration (FHWA) with the aim of facilitating transportation engineers, urban planners, and other practitioners in proactively prioritizing pedestrian safety at crosswalks and intersection approaches based on intersection characteristics.
By utilizing variables that indicate an el- Vol.
12 No.
2 (May 2.
evated risk to pedestrians, the PedISI model enables the identification of crosswalks and intersection approaches that warrant the highest priority for pedestrian safety improvements within a specific study area.
Once high-priority locations are identified, the local authority can conduct more detailed site evaluations to determine the most appropriate safety interventions for each specific context.
The PedISI framework, although widely adopted, is not the only one framework used to assess pedestrian safety.
Micro-level pedestrian safety audits that have a strong connection to the safe system approach have been adopted by European and East Asian countries (Fitzpatrick et al.
, 2024.
Kwon et al.
Rastogi et al.
, 2.
Unlike PedISI, which assesses roadway design and traffic characteristics to improve pedestrian crossing facilities, a safe system approach focuses on forgiving road environments and reducing kinetic energy at points of conflict.
However, this approach requires an extensive behavioral.
The PedISI framework has been used in many countries and contexts.
Avinash et al.
and Zhang et al.
investigated the pedestrian crossing safety index by developing a multiple linear regression model to identify the factors influencing pedestrian safety.
Their findings revealed that the minimum pedestrian safety margin when crossing a roadway depends on parameters such as pedestrian speed, vehicle distance, vehicle speed, pedestrian age, platoon size, pedestrian rolling behavior, vehicle type, and driver yielding behavior.
Although previous studies primarily implemented the PedISI framework in a developed-country context.
PedISI has been implemented in developedcountry context with different traffic characteristics such as in Brazil (Pietrantonio and Fernando Bizerril Tourinho, 2.
and Indonesia.
Studies by Fitriadi .
and Al Rasyid .
examined pedestrian crossing safety in Bandung City.
Indonesia.
Their findings revealed several influential variables, including land use type, crosswalk markings, leg width, availability of traffic signals, and daily traffic volume.
In contrast to previous studies that focused on both four-legs signalized and unsignalized intersections and incorporated user behaviour characteristics as independent variables, this research concentrates specifically on unsignalized three-leg intersections with data focused on pedestrian / userAos perception rather on collecting behavioural data.
User perception data offer several advantages (Ihssian and Ismail, 2.
over behavioral or accident-based data in pedestrian safety analysis .
a Costa et al.
, 2.
Most notably, perception data are easier to collect than accident records, which are often scarce or While behaviour data, such as near miss accident can be obtained by conducting observation study, such methods are resource-intensive Vol.
12 No.
2 (May 2.
and may pose safety risks to researchers (Ogwude et al.
, 2.
In contrast, user perception data provide a safer and more efficient alternative while also capturing subjective assessments of risk (Zhang et al.
, 2019.
Zhu et al.
, 2.
Grounded in risk perception theory, this approach reflects pedestriansAo evaluation of potential danger, enabling analysis of geometric and traffic-related factors independent of actual crash data, i.
, pedestrians may feel that crossing facilities are unsafe even if accident data show low figures (Avinash et al.
, 2.
Nevertheless, the challenge of using perception data to develop a pedestrian safety index model lies in the applicability of the model and its transferability to other sites (Zhu et al.
, 2.
Journal of the Civil Engineering Forum
3 RESEARCH FRAMEWORK AND METHODOLOGY
The core framework of the PedISI model consists of four key steps for assessing pedestrian safety:
crash data analysis, behavioral observations .
ncluding conflicts and evasive maneuver.
, and expert or user-based evaluations.
These elements are conceptualized within a safety assessment hierarchy often illustrated as a pyramid .
ee Figure This paper specifically uses the low-risk approach using ratings based on user opinions.
PedISI was developed using multiple linear regression analysis to establish a relationship between the dependent variableAinamely, the average pedestrian safety scoreAiand the numerical values of independent variables describing intersection geometry, pedestrian facilities, and traffic conditions.
This research follows four main steps (Figure .
: .
Problem identification including questionnaire development based on literature reviews of previous studies dealing with PedISI frameworks.
Data collection including both secondary data and primary surveys such as a pilot survey.
Statistical tests for collected data.
PedISI model estimation and analysis including regression classical assumption tests and sensitivity analysis for policy implementation.
Figure 1.
Safety index measurements Data collection was conducted based on the identified factors that influence intersection safety.
This data can be collected from previous research / official data .
secondary dat.
or collected directly from the field .
primary dat.
The list of data and the collection method are described in the next section.
Following data collection, statistical data testing was conducted, including CronbachAos alpha reliability test and the Pearson correla- Figure 2.
Research Framework Journal of the Civil Engineering Forum Vol.
12 No.
2 (May 2.
tion test.
CronbachAos alpha reliability test evaluates whether multiple questionnaire items intended to measure the same general concept .
, safety perception, satisfaction, anxiet.
do so reliably.
In this paper.
CronbachAos alpha is used to test the reliability of respondent perceptions of the pedestrian crossing safety index.
Meanwhile, the Pearson correlation test is performed for all variables to analyze all variable correlations.
The aim is to evaluate multicollinearity and to help select the most relevant independent variables that are significantly correlated with the dependent variable.
Lastly, the PedISI model is developed using the multiple linear regression method.
The dependent variable .
in the model is the average pedestrian safety rating provided by respondents, while the independent variables .
i ) include the physical characteristics of the intersection and traffic characteristics at the location.
The a and b coefficients need to be estimated.
The multiple linear regression can be mathematically expressed as shown in Equation 1.
y = a b1 x1 .
bi xi Before applying the model to the study locations, the classical assumption test for regression is conducted.
The aim is to make sure the estimated coefficients are unbiased, valid, and dependable.
After the regression model is estimated and tested, the safety level of each leg of the intersection is calculated.
This study utilizes the principle that the lower the safety score of a location, the higher its priority for evaluation.
This is due to the greater potential danger faced by pedestrians at such locations.
By assessing locations with low safety ratings, it becomes possible to identify the contributing factors to the high crash risk and to develop appropriate strategies to improve pedestrian safety.
Figure 3.
Three leg Intersection Study Location .
ource: google ma.
that underlie the pedestrian safety problems in Cimahi City, .
number of intersection legs .
, .
number of lanes per direction .
, 2, and 3 lane.
availability of traffic signals, .
presence or absence of zebra crossings, .
type of land use around the site, .
presence or absence of a median.
5 SURVEY AND DATA COLLECTION
4 CASE STUDY
This paper takes the three-leg intersections in the city of Cimahi as a case study.
Cimahi has many three-leg intersections across the city and more importantly these intersections are unsignalized.
Therefore, it is the ideal case study to obtain enough three-leg intersection facilities as a sample to build user perception data on the pedestrian crossing safety index based on different characteristics of the The approach of using user perceptions and the method for developing the pedestrian safety index model can be useful for replication in other locations, particularly if a post-analysis test of the regression model is conducted to examine its generalizability and predictive power.
In total, fourteen three-leg intersections, in which 42 pedestrian crossing facilities are evaluated .
ee Figure .
The following are the criteria for the important conditions of unsignalized intersections In this paper, 11 variables are considered for developing the PedISI model.
In total, there are three data categories that need to be collected to develop the PedISI model: .
Road geometry .
, traffic characteristics .
, and crossing safety index.
The variables considered and the data collection methods are described in Table 1.
The crossing safety index is collected through a questionnaire survey.
The index uses a Likert scale system, where a value of 1 indicates a very safe pedestrian crossing and a value of 5 indicates a very unsafe pedestrian crossing.
Fourteen questionnaires were prepared, representing fourteen threelegged intersection locations across Cimahi and thus in total forty-two pedestrian crossing facilities were investigated.
In each of these locations, at least 30 respondents were gathered to assess the safety of the intersection.
The location and the number of respondents for each location can be seen in Table 2.
Vol.
12 No.
2 (May 2.
Journal of the Civil Engineering Forum Table 1.
Data requirement and collection method Variable Pedestrian Crossing safety index Traffic volume 85th percentile speed Lane width Number of lanes Maximum speed limit Dominant land use type Markings availability Median width Presence of small alleys On-street parking Collection Method Questionnaire Traffic counting Speed Gun Wheel meter / Roll meter Visual Secondary data Visual Visual Roll meter Visual Visual Description Scale 1 (Very Saf.
- 5 (Very Unsaf.
Vehicle/hour km/hour 1, 2, or 3 30, 50, and 60 km hour-1 0 = Non-commercial.
1 = Commercial 1 = Yes.
0 = No 1 = Yes.
0 = No 1 = Yes.
0 = No Figure 4.
Resume of traffic volume and speed data sure valid responses, ratings were only allowed after the full video was viewed, and each respondent evaluated just one intersection to avoid fatigue.
The video included actual pedestrian crossing the road to increase ecological validity of the responses given.
Random sampling was used, but only Cimahi residents aged 17Ae55Aicapable of independent crossingAiwere included.
A total of 623 respondents participated, with 72% aged 17Ae30 and 28% aged 30Ae55.
Figure 5.
Speed and Traffic volume survey scheme In the survey, the respondents answered questions regarding their perception of the safety index after watching a 60-second video visualising pedestrian crossing conditions and situations.
To en- Regarding road geometric data, the data show that 66% of roads are 2-lane, 2-way undivided .
/2 UD), 14% are 4-lane, 2-way divided .
/2 D), 14% are 2lane one-way .
/1 UD), and only 4% are 4-lane, 2way undivided .
/2 UD), indicating that only 14% of roads have a median.
Regarding lane width, 85% of roads have 2.
5 m lanes, 14% have 3.
5 m, and only 2% have 5 m lanes.
Figure 4 shows summary of speed and traffic volume data at the 14 study locations.
Traffic data, including volume and speed, were col- Journal of the Civil Engineering Forum Vol.
12 No.
2 (May 2.
Table 2.
Number of observations for safety index data Intersection Cihanjuang Pemkot Cihanjuang Kecamatan (Perpustakaa.
Kecamatan Pesantren Aruman Pesantren Sangkuriang Kamarung Encep Kartawirya Raden Demang Hardjakusumah Jend.
Sudirman Dustira Gandawijaya 2 Pojok Utara H Amir Mahmud Total Leg West North West East North West South East North South West North South East West South North North South West West East North North South East East North South West East North West East South South West East South West East North Respondent Observation lected using a speed gun at unsaturated conditions near intersections, both for approaching and departing vehicles .
ee Figure .
The speed data were then analysed based on the average and the 85th
6 RESULTS AND DISCUSSION
Firstly, we analyzed the respondentsAo answers to the safety rating using the CronbachAos Alpha reliability test for each respondent group.
Table 3 presents the results of the data reliability test for each respondent group, where each group represents an intersection.
The results show that all data sets have CronbachAos Alpha values () above 0.
7, indicating high reliability.
Next.
Pearson correlation test was used to detect relationships among variables in the linear model.
Variables with high correlation coefficients .
reater than 0.
were excluded from the Table 4 shows the results of the Pearson correlation.
Variable X4 has a high correlation with variables X5 and X8 , and thus variable X4 was eliminated from the model due to its strong correlation with two other independent variables.
Table 5 shows the results of the PedISI regression We conducted several iterations to ensure the best model for further analysis.
The results reveal that pedestrian crossing safety is significantly influenced by traffic volume, vehicle speed, lane width, and median width.
Across all five model iterations, three variables consistently showed strong and statistically significant effects: traffic volume (X1 ), 85th percentile speed (X2 ), and lane width (X3 ).
Higher traffic volume and vehicle speeds are associated with lower safety ratings, indicating that greater vehicle exposure and speed pose increased risks to pedestrians.
Similarly, wider lanes tend to reduce safety perception, likely due to longer crossing distances and the possibility of vehicles traveling faster in wider lanes.
Conversely, median width (X8 ) was a significant positive contributor to safety perception among respondents in the model.
wider median improves pedestrian safety by offering a safe waiting zone, especially on multi-lane roads.
Other variablesAisuch as speed limit signage, onstreet parking, and land useAiwere found to be statistically insignificant and were gradually removed to refine the model.
These findings align with previous studies on midblock and three-leg pedestrian crossings, which concluded that vehicle speed, traffic volume, and road geometric consistency significantly influence pedestrian crossing safety (Avinash et al.
, 2019.
da Costa et al.
, 2.
This consistency arises from similar conditions in which pedestrians must cross without signalized aid, such as pelican crossings.
In such cases, unlike signalized intersections where delays are the primary factors influencing pedestrian crossing behavior and thus safety, at unsignalized facilities pedestrian safety is more affected by traffic volume and speed (Hashemi et al.
, 2.
The regression model was developed iteratively, with attention to the modelAos goodness of fit, the predictive power of the model, and the significance of the estimated coefficients.
The goodness of fit of a regression model evaluates how well the model estimates actual values.
Statistically, it can be assessed through the coefficient of determination (R2 ) and the F-statistic.
The predictive power of the model is evaluated using the Root Mean Square Error between the data and the predictive value of the model.
Meanwhile, the statistical significance of the independent variablesAo estimated parameters is evaluated using the t-test or p-values (Ghozali.
Vol.
12 No.
2 (May 2.
Journal of the Civil Engineering Forum Table 3.
CronbachAos Alpha Reliability Test Results Set CronbachAos N observation Description High Reliability High Reliability High Reliability High Reliability High Reliability Medium Reliability High Reliability Set CronbachAos N observation Description High Reliability High Reliability High Reliability High Reliability High Reliability High Reliability High Reliability Table 4.
Results for Pearson Correlation Test
X10
X10
Table 5.
Results for PedISI Regression Model
Constant
X10
Adjusted R2
RMSE
F-test X1 :
X2 :
X3 :
X4 :
X5 :
Iteration 1 Iteration 2 Coeff P-value Coeff P-value Pedestrian crossing safety rating Traffic Volume .
85 percentile Speed (Kp.
Lane width .
Number of lane Max Speed Limit (Kp.
Iteration 3 Coeff P-value X6 :
X7 :
X8 :
X9 :
X10 :
The regression analysis of the collected pedestrian crossing perception data in this study demonstrates that the model has strong predictive power, explaining 75Ae76% of the variance in pedestrian safety ratings across sites, with only a modest increase in prediction error as the model is simplified.
The final model, with only four significant predictors, provides the best balance between model simplicity and explanatory power (Adjusted R2 = 0.
RMSE
= 0.
Despite using fewer variables, the final Iteration 4 Iteration 5 Coeff P-value Coeff P-value Dummy for land use, 1 if commercial otherwise 0 Dummy for road marking, 1 if any otherwise 0 Median width .
Dummy for small alley at intersection, 1 if any otherwise Dummy for on street parking, 1 if any otherwise 0 model remains statistically robust and nearly as accurate, making it effective for comparing and predicting safety across different urban crossing sites.
These statistical test results for the model show that while perception data introduces subjectivity, it can be made predictive and generalizable if it is collected systematically, linked to real-world measurable features, and modeled using robust statistical Journal of the Civil Engineering Forum Vol.
12 No.
2 (May 2.
Figure 6.
Sensitifity Analysis Table 6.
Heteroscedasticity Test and Multicollineraity test Results Variabel Glejser p- VIF value X1 = Traffic Volume X2 = 85th Percentile 0.
Speed X3 = Lane Width X8 = Median Width To ensure external validation of the regression model, a cross-validation test was performed.
Specifically, the Leave One Out Cross Validation (LOOCV) technique was employed, particularly due to the small data.
LOOCV is a special case of k-fold cross-validation where k equals the number of observations (N ) (Kuh et al.
, 2.
For this case, the LOOCV analysis ran 42 regressions, each time training the model on 41 observations and testing on the one that was left out.
This process helps estimate how the model would perform on unseen data.
The result showed a Root Mean Square Error (RM SE) of This means that, on average, the modelAos prediction of pedestrian crossing safety ratings was off by about 0.
43 units .
n the safety rating scal.
, when applied to unseen data .
ne site at a This is a relatively low error, considering that the rating scale, i.
the independent variable, appears to range roughly between 1.
6 to 4.
8, indicating good generalizability.
In summary, despite relying on subjective perception data, this study demonstrates that with systematic data collection and robust statistical methods, it is possible to develop a regression model with strong predictive power and Further, to ensure a robust ordinary least square (OLS) regression, classical assumption tests were conducted on the selected model.
Classical assumption testing is a series of statistical tests conducted to ensure that a multiple linear regression model satisfies the underlying statistical assumptions.
These assumptions are essential to guarantee the accuracy of the parameter estimates generated by the model (Nugroho et al.
, 2.
The classical assumption tests include: .
Normality test, .
Autocorrelation test, .
Heteroscedasticity test, and .
multicollinearity test.
The results of classical assumption tests are as follows:
A The normality test aims to determine whether the residuals are normally distributed.
This test is conducted using a standard KolmogorovSmirnov test.
The result of the normality test on the model residuals using the KolmogorovSmirnov test showed a significance value of Since this value is greater than 0.
05, it can be concluded that the residuals are normally distributed (Priyatno, 2.
A Autocorrelation refers to a condition in a linear regression model where there is a dependency among the residuals .
, the differences between actual and predicted value.
across successive observations.
One common method to test for autocorrelation is the Durbin-Watson (DW) test.
The DW value of our regression model is 2.
077, which is greater than the lower bound du value of 1.
720 and less than the upper bound value of 6.
This result implies that there is no autocorrelation in the regression model.
A The heteroscedasticity test is conducted to examine whether there is a variance inequality in the residuals across observations in the regression model.
If the residual variance remains constant across observations, the condition is called homoscedasticity.
if it varies, it is referred to as heteroscedasticity.
A good regression model is one that exhibits homoscedasticity or does not suffer from heteroscedasticity.
The Glejser test is performed to test the heteroscedasticity of the model.
Table 6 shows Vol.
12 No.
2 (May 2.
Journal of the Civil Engineering Forum the heteroscedasticity test for regression independent variables using the Glejser test.
The results show that all variables have p-values greater than 0.
05, implying that there is no heteroscedasticity in the model.
A The multicollinearity test is conducted to detect the presence of high correlations among independent variables in a regression model.
VIF is used to test multicollinearity in the regression model.
VIF is the reciprocal of tol- erance .
e V IF = T olerance ), in which Tolerance is defined as 1 Ae R2 , where R2 is the coefficient of determination obtained by regressing one independent variable on all the others.
Table 6 shows the VIF values, and the results show that the VIF value ranges between 1-5, which indicates moderate correlation but is usually acceptable (Priyatno, 2.
Table 7.
Results of Model Analysis of Road Crossing Safety Values in Cimahi City No Intersection Cihanjuang Pemkot Ae Cihanjuang Kecamatan (Perpustakaa.
Kecamatan Ae Pesantren 12 Aruman 15 Pesantren 18 Sangkuriang 21 Kamarung 24 Encep Kartawirya 27 Raden Demang 28 Hardjakusinah 30 Jend.
Sudirman 33 Dustira
36 Gandawijaya 2
37 Pojok Utara
40 H Amir Mahmud
Average Standard Deviation Leg Direction West
North
West
East
North
West
South
East
North
South
West
North
South
East
West
South
North
North
South
West
West
East
North
North
South
East
East
North
South
West
East
North
Barat
East
South
South
West
East
South
West
East
North
Index
Journal of the Civil Engineering Forum
7 MODEL IMPLEMENTATION AND SENSITIVITY ANALYSIS
In this section, the regression equation is applied to calculate the pedestrian safety index at the study Generally, the results show that the three-legged intersections in Cimahi have adequate pedestrian crossing safety index with 83% of the locations studied having higher ratings above 2.
However, there are seven locations where the index values are below 2.
5 indicating safety risks for Table 7 shows the overall index value for all study locations.
Jalan Raya Cibabat .
est leg of Pesantren intersectio.
has the lowest safety index with pedestrian crossing safety index value of This low safety index implies an urgent need to improve the situation at this location.
The low index value is caused by the high value of 85th percentile speed in which reaches 43 km/h, the highest among the study location areas.
In addition.
Jalan Raya Cibabat Barat, which consists of four lanes, each with a lane length of 4 meters and a narrow median, has one of the highest traffic volumes in Cimahi CityAireaching 1,130 vehicles per hour.
These conditions create a hazardous crossing environment.
Pedestrians must interact with high-speed vehicles over an extended period due to the long crossing Vol.
12 No.
2 (May 2.
duction, reducing vehicle volume is another strategy to improve the pedestrian safety index at intersections.
This can be achieved by converting certain lanes into dedicated pedestrian, bicycle, or public transport lanes .
BRT or other transit mode.
Such interventions promote a shift toward public transportation, reduce private vehicle usage, and enhance safety and comfort for non-motorized road Redesigning road geometry such as reducing lane width and increasing median width, can also contribute positively to improving safety.
Nevertheless, these factors must be approached with caution.
Lane width and median width are subject to road design standards that ensure road functionality and vehicle flow.
Excessive narrowing of lanes may compromise vehicle maneuverability, and median expansion is only practical up to a reasonable limitAi an excessively wide median .
, 100 .
would be infeasible and counterproductive in an urban context.
Therefore, adjustments to lane and median widths should be considered only within the limits of established design standards.
8 CONCLUSIONS
Next, a sensitivity analysis was conducted by gradually adjusting the values of the independent variables and observing the resulting changes in the dependent variable.
Each independent variable was modified by 5% from its initial value, while for locations without a median, the adjustment was made by increasing the width by 5% of the maximum median width.
Figure 6 presents the results of the sensitivity analysis, in which the values of the independent variables were altered incrementally by The analysis shows that a reduction in the 85th percentile vehicle speed exhibits the highest sensitivity compared to changes in other independent Specifically, each 5% reduction in the 85th percentile speed results in a 6% increase in the safety index of the pedestrian crossing location.
The pedestrian safety index model for unsignalized three-leg intersections in Cimahi City is formulated based on regression analysis.
The advancement of this study lies in its focus on unsignalized three-leg intersections, a junction type often overlooked in pedestrian safety research.
Unlike typical mid-block or four-leg intersections, three-leg intersections present unique geometric and behavioral challenges due to asymmetric traffic patterns and irregular pedestrian paths.
The use of a perceptionbased safety rating, derived from video stimuli and validated through rigorous statistical testing, offers a novel methodological approach that captures realworld pedestrian perception of the safety of pedestrian crossings.
This approach bridges subjective safety perceptions with objective road characteristics, enabling more holistic and human-centered safety assessments.
The results indicate that speedmanagement should be prioritized to improve perception of pedestrian safety, as reducing vehicle speed provides greater improvements than any other intervention.
Efforts to reduce vehicle speed and enhance pedestrian safety through increased visibility and the provision of safe refuge areas can include the following measures: elevating the roadAos median, raising pedestrian crossing installing pedestrian crossing islands, and elevating intersection platforms.
Additional speed-reducing treatments include rumble strips, speed tables, speed humps, chicanes .
urved road alignment.
, and road narrowing .
hokers or pinch point.
Aiparticularly in areas prone to crashes or with high pedestrian activity.
Besides speed re- The pedestrian safety index at 42 unsignalized three-leg intersections in Cimahi ranges from 1.
5 to 5, with an average score of 3.
17 and a standard deviation of 0.
Accordingly, these values fall within the Aumoderately safeAy category for pedestrian crossings.
The analysis shows that traffic vehicle volumes, traffic speeds, lane width, and median width are significant factors affecting pedestrian crossing safety perceptions.
Analyzing the model, the R2 value of 0.
78 indicates that the model explains 78% of the variation in safety index scores.
This suggests that the model is sufficiently accurate in predicting safety levels based on vehicle volume, speed, lane width, and median width.
The model also produces Vol.
12 No.
2 (May 2.
good predictive power, indicated by an RMSE value The strong predictive power and generalizability of the model make it a powerful tool for local governments to quantify and respond to public safety concerns, using predictions to prioritize significant interventions even in areas where direct perception surveys are not feasible.
The model results align with previous studies, which found that traffic volume, traffic speed, lane width, and median width are significant factors for unsignalized crossing facilities (Rastogi et al.
, 2011.
Zhu et al.
From the sensitivity analysis, the 85th percentile speed was found to be the most influential variable compared to the other independent variables.
Therefore, reducing vehicle speed is the most recommended strategy to improve pedestrian crossing safety.
These findings support infrastructurefocused interventions such as lane narrowing, speed management, and median installation to enhance pedestrian safety particularly for three leg intersections facilities .
a Costa et al.
, 2.
This study has several limitations that should be acknowledged.
First, traffic volume and speed data were collected only during the peak hour observation period, which may not fully capture daily or temporal variations in intersection conditions.
Similarly, the video used to represent the intersections reflects only an ideal situation, recorded during calm daytime weather.
Pedestrian perceptions may differ significantly under alternative conditions such as nighttime, rainfall, or other adverse These limitations suggest that the ecological validity of the findings could be improved in future research.
One promising approach is the use of virtual reality (VR) experiments to simulate diverse crossing environments and conditions, enabling the assessment of pedestrian perceptions under more realistic and varied scenarios.
Future studies may also explicitly test the impact of safety measuresAisuch as lane width reduction, stricter speed control, or median adjustmentsAiusing VRbased experimental designs.
Additionally, incorporating individual pedestrian characteristics, such as trip purpose, age, and gender, alongside contextual conditions like weather and lighting, would provide a more comprehensive understanding of perception-based pedestrian safety.
DISCLAIMER
The authors declare no conflict of interest.
REFERENCES