Predicting Freeway Travel Time Using Multiple- Source Heterogeneous Data Integration

Round 1

Reviewer 1 Report

The paper presents a method of estimating travel time through SVM using other variables such as accident, day of the week.  The paper needs an extensive revision in its English before getting considered for the publication. Some questions/ comment:

Is  GPS source of data is used as one of the variables to estimate the travel time? There are some gaps between prediction and actual values. What if GPS data is used for estimation of the actual travel time? Can you plot it along with actual and predicted values? 

Author Response

Point 1:

The paper presents a method of estimating travel time through SVM using other variables such as accident, day of the week.  The paper needs an extensive revision in its English before getting considered for the publication.

Response 1:

Thanks for your suggestion.

We invited professor Lee.D. Han to check the theoretical content and grammar of the whole manuscript. Professor Han is from the University of TENNESSEE in USA, in addition, his research area is civil engineering.

Also, we invited another English speaking editor Lisa to correct and "light" the whole word of the paper, Lisa is from the University of TENNESSEE in USA.

Point 2:

Is  GPS source of data is used as one of the variables to estimate the travel time? There are some gaps between prediction and actual values. What if GPS data is used for estimation of the actual travel time? Can you plot it along with actual and predicted values? 

Response 2:

I am sorry that we did not use GPS data for estimation in this paper.

Compared with GPS data, toll data is more than correct which can record every vehicle entry time and leaving time. In fact, GPS data can be considered in the further study, although it has some shortage when used for travel time prediction . Thank your suggestion.

Reviewer 2 Report

In the paper real data are analyzed and travel time functions are estimated. The paper has the merit to use real data. The estimated models are not compared with consolidated models reported in literature.

1. The point where equation (1) is considered in the model is not reported. Sum of terms with different unit of measurement are considered. The procedure for the calibration of the value reported (3/6, 2/6, 1/6) is not described.

2. In the travel time forecast, consolidated behavioral models proposed in literature are not considered (i.e. Ben-Akiva et al., 2001, Networks and Spatial Economics; Mahmassani, 2001, Networks and Spatial Economics; Alonso et al., 2017 Transportation Research Procedia; Chilà Giovanna et. al., 2016, IET Intelligent Transport Systems). The proposed approach could be better than existing consolidated approaches, but a comparison (at least in the state of the art section) is needed, highlighting pros and cons for each one.

3. The main variable that influences the travel time prediction is the traffic flow. The flow is not reported among the variables of table 3. The influence of time from traffic flow can be considered.

4. The values adopted for the parameters considered in section 4.1 have to be justified.

5. For the numeric interpretation, the calibrated weight for the input and output variables and the relative statistic indicators are required. In the numeric results, only the errors are reported. The interpretation of each considered input and output variable could help for the interpretation of the phenomena. Errors allow comparing only the tested models.

Minor

Check if for a figure the authorization of using it could be required.

Table 2 is divided in two pages.

Figures and tables can be linked in the text.

Nuclear parameter is not defined.

Author Response

Point 1:

In the paper real data are analyzed and travel time functions are estimated. The paper has the merit to use real data. The estimated models are not compared with consolidated models reported in literature.

Response 1:

Thank your appreciation and suggestion. In this paper, we proposed one travel time predicting model based on multiple source data. To prove its accuracy, we compared the new model with BP neural network model and common SVM. The reason why we selected BP and SVM as comparison is that these two model are more than commonly used. Anyway, other models reported in literature can be considered in the further study.

Point 2:

The point where equation (1) is considered in the model is not reported. Sum of terms with different unit of measurement are considered. The procedure for the calibration of the value reported (3/6, 2/6, 1/6) is not described.

Response 2:

I am sorry that the explanation for equation (1) is not clear in our paper. In the revised version, we have improved it.

Also ,we revised equation (1), undefined numbers (3 / 6, 2 / 6, 1 / 6) were no longer used.

Point 3:

In the travel time forecast, consolidated behavioral models proposed in literature are not considered (i.e. Ben-Akiva et al., 2001, Networks and Spatial Economics; Mahmassani, 2001, Networks and Spatial Economics; Alonso et al., 2017 Transportation Research Procedia; Chilà Giovanna et. al., 2016, IET Intelligent Transport Systems). The proposed approach could be better than existing consolidated approaches, but a comparison (at least in the state of the art section) is needed, highlighting pros and cons for each one

Response 3:

Thank your suggestion. In the revision ,we cited these research, and listed in the literature (Table 1) and references, also, we highlight the pros and cons of these methodologies in the section 1.2.

Point 4:

The main variable that influences the travel time prediction is the traffic flow. The flow is not reported among the variables of table 3. The influence of time from traffic flow can be considered.

Response 4:

I am sorry that the manuscript did not clearly list the variable "traffic flow" in the travel time prediction model. Actually, we introduced two other variables "holiday" and "weekday" to replace "traffic flow". Anyway, traffic flow is truly appropriate variable ,will be considered in the further study.

Point 5:

The values adopted for the parameters considered in section 4.1 have to be justified.

Response 5:

Many thanks for your suggestion. We supplemented the details about the parameters value determination, listed as line 572 to 578.

Point 6:

For the numeric interpretation, the calibrated weight for the input and output variables and the relative statistic indicators are required. In the numeric results, only the errors are reported. The interpretation of each considered input and output variable could help for the interpretation of the phenomena. Errors allow comparing only the tested models.

Response 6:

Thank your kind suggestion. For the numeric interpretation, we supplemented some details in 4.2.

For the phenomena interpretation, we conducted sensitivity analysis in section 4.3, and revealed the effect of model inputs.

Point 7:

Check if for a figure the authorization of using it could be required.

Table 2 is divided in two pages.

Figures and tables can be linked in the text.

Nuclear parameter is not defined.

Response 7:

Thank your kind suggestion. We improved these in our revised version.

Round 2

Reviewer 1 Report

Thanks to authors for addressing my comments.

Author Response

Point 1:

English language and style are fine/minor spell check required .

Response 1:

Thanks for your suggestion.

We "light" the word for the whole paper.

Reviewer 2 Report

Response 3.

In the reference number 22 the names and surnames are inverted (Chilà G.; Musolino G.; Polimeni A. instead of Giovanna C.; Giuseppe M.; Antonio P.).

In the same new reference number 22 some authors’ names are not reported (Rindone C.; Russo F.; Vitetta A.).

In the new reference number 23 some author’s names are not reported (Rindone C.; Vitetta A.).

In some cases, in the references a space between names is required.

Check the reference number in the text (i.e. lines 127, 129).

Response 5.

The lines 572-578 are not in the paper. The paper ends at line 540.

Response 6.

The calibrated weight for the input and output variables and the relative statistic indicators could increase the information reported in section 4.3.

Author Response

Point 3:

In the reference number 22 the names and surnames are inverted (Chilà G.; Musolino G.; Polimeni A. instead of Giovanna C.; Giuseppe M.; Antonio P.).

In the same new reference number 22 some authors’ names are not reported (Rindone C.; Russo F.; Vitetta A.).

In the new reference number 23 some author’s names are not reported (Rindone C.; Vitetta A.).

In some cases, in the references a space between names is required.

Check the reference number in the text (i.e. lines 127, 129).

Response 3:

Thank your suggestion. In the revision ,we cited these research, and listed in the literature (Table 1) and references, also, we highlight the pros and cons of these methodologies in the section 1.2.

Point 5:

The lines 572-578 are not in the paper. The paper ends at line 540..

Response 5:

Many thanks for your suggestion. We supplemented the details about the parameters value determination, listed as line 572 to 578.

Point 6:

For the numeric interpretation, the calibrated weight for the input and output variables and the relative statistic indicators are required. In the numeric results, only the errors are reported. The interpretation of each considered input and output variable could help for the interpretation of the phenomena. Errors allow comparing only the tested models.

The calibrated weight for the input and output variables and the relative statistic indicators could increase the information reported in section 4.3.

Response 6:

Thank your kind suggestion. For the numeric interpretation, we supplemented some details in 4.2.

For section 4.3, we made major revision. We added some calculations in section 4.3.1 and 4.3.2, and supplied two figures in these two sections.