AI-Powered Evaluation of On-Demand Public Transport: A Hybrid Simulation Approach

Round 1

Reviewer 1 Report

Comments and Suggestions for Authors

This study presents an interesting framework for evaluating on-demand public transport using a BiLSTM-based demand model. The real-world data foundation makes the work timely and policy-relevant, but several improvements are needed to strengthen its academic contribution. Please consider the following comments.

1. The introduction outlines the need for on-demand transit, but it should be supported with stronger quantitative evidence (e.g., off-peak demand declines, operating cost increases for fixed routes). Clearer statistics on why existing services underperform would make the motivation more compelling.
2. Related work on AI-based skip-stop or express-service estimation should be reviewed, including studies such as “Estimating Express Train Preference… using XGBoost” and “Loading and Unloading Time Estimation… using XGBoost.”
3. Although BiLSTM forecasting is central to the framework, the distinctions from LSTM/GRU/TCN models and the rationale for choosing BiLSTM are not well discussed. More detail is needed on input windows, seasonality handling, route-level heterogeneity, and overfitting checks.
4. The Commuter-based ABM is well described, but key parameters (e.g., generalized cost coefficients, walking speed, wait-time penalties) lack justification or calibration details. Given the impact of skipped-stop logic and depot catchment settings, sensitivity analysis is essential.
5. While Scenario 2 shows clear performance gains (e.g., 26–32% travel time reduction, 72% CO₂ reduction), these results may be specific to Melbourne’s network. More discussion is needed on generalizability to other contexts (e.g., high-density vs low-density cities) and on institutional or financial implications.
6. The study applies SP-based adoption rates (17% in peak, 100% off-peak), but SP–RP gaps and the limited, localized sample (n=327) raise concerns about representativeness. Without explicit utility formulations or sensitivity analysis, the validity of mode-shift assumptions remains uncertain; more rigorous behavioural modelling (e.g., logit, mixed logit, nested logit) would strengthen this component.

Author Response

Please see attached

Author Response File: Author Response.pdf

Reviewer 2 Report

Comments and Suggestions for Authors

In this article, the authors present research results on a hybrid simulation framework that integrates deep learning-based demand forecasting, behavioral research data, and agent-based simulation to assess the performance of an on-demand public transport system at both the operational and user levels. The case study was Melbourne, Australia. 
Below are my comments that may help improve the article:
1. I suggest supplementing the abstract with a detailed presentation of the authors' research results, including specific numerical values.
2. The discussion requires further clarification. It doesn't include comparisons with results obtained by other researchers.
3. Please also indicate the limitations of the conducted research.
4. The text contains references to figures without providing their numbers. Please review the entire text and specify which figure is being discussed by providing its number.
5. When using an abbreviation for the first time, please provide its expansion, e.g. BiLSTM

Author Response

Please see attached

Author Response File: Author Response.pdf

Reviewer 3 Report

Comments and Suggestions for Authors

This paper described a framework for evaluation of on-demand public trasport systems performance based on combination of deep-learning-based demand forcasting, behavioral survey data, and agent-based simulation.

The structure of the paper is acceptable. It is nice that the related work is placed in a separate section. However, I would recommend to avoid 4th-level headings in Section 3.1.3 (names of the modules), as 4th-level headings are discouraged even in longe texts. Remove these headings completely, or change structure to enable usage of 3rd-level headings.  

Individual parts of the framework are described in detail. For the agent-based simulation, the non-standard modules, which are required specifically for this framework could be described in more detail, especially the Dynamic Dispatch Scheduling Module. The demand prediction and public transport services modeling seems to be described in more than sufficient detail.

The case study uses two scenarios and baseline, which is enough to sufficiently demonstrate the ability of the proposed framework to asses the performance of on-demand public transport systems.

The figures are adequate and of sufficient quality.

The references are relevant and up-to-date.

Comments on the Quality of English Language

The English is very good, errors and typos are rare. Still, proofreading by a grammar-skilled native speaker can further improve the paper.

Author Response

Please see attached

Author Response File: Author Response.pdf

Round 2

Reviewer 1 Report

Comments and Suggestions for Authors

I am happy with the responses

Reviewer 2 Report

Comments and Suggestions for Authors

Thank you for your replies. I have no comments.