Pith. sign in

REVIEW 1 major objections 4 minor 17 references

An Embarrassingly Simple Rule-based Visiting Circulation Approach to Trip Destination Prediction

T0 review · 1 major / 4 minor · reviewed 2026-08-01 · deepseek-v4-flash

Challenge this review Re-run · record.json Download PDF Read on arXiv ↗

Pith's one-line read A single rule—the destination of each trip is the origin of the next—predicts trip destinations at 43.8% accuracy, outperforming supervised baselines and ranking second in the competition.

desk verdict A simple deterministic rule — destination of trip i equals origin of trip i+1, with the last looping to the first — earned second place on a real leaderboard; the paper is honest about its limits and worth a serious referee, though the claim that it significantly outperforms supervised learning is overstated. read the letter →

arxiv 2607.25751 v1 pith:B7XYHOWJ submitted 2026-07-28 cs.LG

Eng-Shen Tu , Yong-Han Chen , En-Chao Liu , Hao-Yun Keng , Cheng-Te Li This is my paper
classification cs.LG
keywords tripdestinationpredictionrule-basedmodelvisitingcirculationroundassumptionorigin-destinationrelationpersonsurveypeopleflowdataIEEEBigCup
verification ladder T0 review T1 audit T2 compute T3 formal

The pith

A machine-rendered reading of the paper's core claim, the machinery that carries it, and where it could break.

The reading

This paper tries to prove that trip destinations can be predicted without any machine learning, using a simple rule based on how people move in a day. The rule, named Rule-based Visiting Circulation (RVC), assumes each person's daily trips form a round trip: for a person with multiple trips, the destination of each trip is the origin of the next trip (sorted by departure time), and the last trip loops back to the first origin; for a single-trip person, the destination is the origin itself. On the Kinki test set, this rule achieves 0.43838 accuracy, far above LightGBM (0.13020), Random Forest (0.23124), and the competition baseline (0.20879). A sympathetic reader would care because it suggests that a basic behavioral regularity—people revisit places and return home—carries more predictive signal than rich features and supervised learning for this task.

What carries the argument

The central object is the Rule-based Visiting Circulation (RVC) rule. For each person ID, trips are grouped and sorted by departure time. For a person with more than one trip, the predicted destination of trip i is the origin of trip i+1, and the destination of the last trip is the origin of the first trip. For a person with exactly one trip, the predicted destination is the origin of that trip. This rule directly encodes the round-trip assumption and the revisit behavior observed in the data.

What would settle it

Take a known dataset of daily trips with true destinations, compute the proportion of people whose trips form a single closed loop (each trip's destination equals the next trip's origin, and the last equals the first origin); if that proportion is substantially below the accuracy RVC achieves (0.43838 on Kinki), the rule's success on Kinki must be partly accidental, and measuring RVC's accuracy on a population where multiple round trips or missing trips are common should show it falling to near the origin-baseline level.

Watch

Extended reading notes

Core claim

The paper's central claim is that individuals' daily trips exhibit a 'visiting circulation' pattern: the destination of one trip is the origin of the next, and the day's travel forms a closed loop. By encoding this as a deterministic rule—sort trips by departure time and chain them, looping the last trip's destination to the first trip's origin—the method predicts destinations in a metropolitan area it has never seen, with no training and no labeled destinations. The rule achieves 0.43838 accuracy on the Kinki area, ranking second in the competition and outperforming supervised models by a wide margin. The paper also demonstrates robustness by applying the same rule to four other metropolita

Load-bearing premise

The model assumes that for each person, the dataset contains every trip made that day and that these trips form one complete round trip, so that departure-time sorting recovers the true stop order and the last destination returns to the first origin.

Editorial extensions

If this is right

  • If daily trips are largely round trips, then origin sequences alone can predict destinations at an accuracy that supervised models using age, gender, occupation, trip type, and zone features cannot match.
  • RVC can generate pseudo-labels for the target metropolitan area, and combining it with supervised models improves their accuracy (e.g., Random Forest rises from 0.23124 to 0.37798).
  • The rule generalizes across metropolitan areas without retraining, as shown by consistent accuracy gains over the baseline in Tokyo, Chukyo, Kyushu, and Higa.
  • The proportion of multi-trip individuals in an area correlates with RVC accuracy, implying that areas with more multi-stop travel are more predictable under this rule.
  • The competition baseline 'destination equals origin' is a special case of RVC for single-trip individuals; RVC's gain comes from chaining multiple trips.

Reading between the lines

Editorial extensions of the paper, not claims the author makes directly.

  • The rule's success suggests that temporal sequencing—the order of a person's stops—is a stronger predictor than contextual features, and future destination-prediction models should incorporate this as an inductive bias rather than relying solely on feature engineering.
  • A testable extension is to segment each person's day into multiple round trips (e.g., home-to-work-to-home, then home-to-shop-to-home) and apply RVC per segment; the paper explicitly identifies the inability to handle multiple round trips as a limitation.
  • The completeness assumption—that the dataset contains all trips a person made that day—is checkable on the four training areas where ground-truth destinations exist; if the chaining accuracy in those areas degrades when trips are artificially removed, it would quantify the rule's sensitivity to missing data.
Share X Bluesky LinkedIn Reddit HN

Editorial analysis

A structured set of objections, weighed in public.

Desk editor's note, referee report, and a circularity audit.

Referee Report

1 major / 4 minor

Summary. The paper proposes the Rule-based Visiting Circulation (RVC) method for the IEEE Big Data Cup 2022 trip destination prediction task. RVC assumes that each individual's daily trips form a round trip. For each person ID, trips are sorted by departure time; the destination of each trip is set to the origin of the next trip, and the destination of the last trip loops back to the origin of the first. Single-trip PIDs are predicted to end at their origin. The method uses no labels from the target metropolitan area and has no learned parameters. The authors report a leaderboard accuracy of 0.43838 on the Kinki test set, ranking second, compared with 0.20879 for the competition baseline, 0.13020 for default LightGBM, and 0.23124 for default Random Forest. They also report offline accuracy on the four training areas after removing destinations, with values ranging from 0.436 to 0.692.

Significance. If the reported leaderboard result is accurate, the paper demonstrates a striking empirical finding: a deterministic rule using only trip ordering can outperform supervised learning approaches in a zero-shot cross-city setting. The method is fully deterministic, has no free parameters, and the central claim is externally verifiable through the competition leaderboard. The paper also includes an honest discussion of limitations and an ablation study. The contribution is empirical rather than methodological, but it is a valuable data point for the destination-prediction community and supports the reproducibility of competition findings.

major comments (1)
  1. [Abstract / IV-A / Table I] The abstract and conclusions claim that RVC 'significantly outperform[s] supervised learning methods' (Abstract, Section V). The only supervised baselines are LightGBM and Random Forest with default hyperparameters (Section IV-A). With no hyperparameter tuning or attempt to adapt them to the cross-city setting, the comparison does not support the word 'significantly.' The paper itself acknowledges this in Section IV-A ('the performance of the methods implemented only reflects the value of these methods in this competition task'). Please either temper the claim (e.g., 'outperforms default-hyperparameter LightGBM and Random Forest') or expand the baseline comparison.
minor comments (4)
  1. [II-A / Table III] The dataset statistics are internally inconsistent. For Chukyo, 113,820 / 343,752 = 0.331, but Table III reports a proportion of 0.554. For Kinki, the text says 34,496 total trips, but Table III implies approximately 967,000 trips (548,834 / 0.5673). Please correct these numbers; the Kinki trip count appears to have been copied from Higa.
  2. [Table V] The description of 'Partial RVC' says 'others set to 0.' This is ambiguous: in Section III-C, single-trip destinations are set to the origin. If '0' is a dummy zone, the ablation is not a meaningful comparison. Please clarify whether single-trip PIDs were assigned to the origin or to a constant zero destination.
  3. [IV-B] Please state explicitly how the supervised baselines were trained (e.g., on which training areas, with what train/validation split) and evaluated on Kinki. The preprocessing is described in detail, but the training procedure is not.
  4. [IV-C] The phrase 'We can obviously see' (Section IV-C) is informal; rephrase to a direct statement such as 'The results show that RVC outperforms the baselines.'

Circularity Check

0 steps flagged · score 0.0 of 10

No significant circularity: RVC is a parameter-free deterministic rule evaluated against external ground-truth leaderboard results; its round-trip assumption is an acknowledged limitation, not a fitted target.

full rationale

The central claim of the paper is empirical: the RVC rule achieves 0.43838 accuracy and second place on the IEEE Big Data Cup 2022 leaderboard. No parameters are fitted and no target destinations are used to construct the rule; the rule deterministically maps each trip's destination to the origin of the next trip (or the same origin for single-trip PIDs) using only provided origin and departure-time information. This prediction is then compared against real held-out destinations, which are external to the method's inputs. The round-trip assumption is explicitly stated in Section III ('we make the assumption that every individual's daily trip is a round trip') and its failure modes are honestly acknowledged in Section III-E, but an assumption is not circularity. The cited references, including the paper by co-author Cheng-Te Li [9], are motivational rather than load-bearing: the rule's validity is established by its leaderboard and offline accuracy, not by the citation chain. The variants in Section IV are ablations and extensions, not attempts to conceal fitted inputs as predictions. Overall, the derivation chain is self-contained with respect to circularity: the result does not reduce to its inputs by construction.

Assumptions & free parameters 0 free parameters · 4 assumptions · 0 invented entities

The method introduces no fitted numerical parameters and no new entities. Its validity rests entirely on the domain assumptions that daily trips are complete single round trips and that this behavior transfers across metropolitan areas.

assumptions (4)
  • domain assumption Each person's trips in a day form a single complete round trip; the destination of the last trip is the origin of the first.
    Stated in Section III and used in the default RVC; if false (multiple round trips, missing trips), predictions break. The paper itself acknowledges this in Section III-E.
  • domain assumption The dataset contains all trips made by each PID within the day.
    Section III-D-1 assumes dataset completeness; if any trip is missing, the chain is wrong.
  • domain assumption Sorting trips by departure time recovers the true stop order, and all trips occur on the same day.
    Section III-A uses departure-time sorting as the only ordering signal.
  • domain assumption Round-trip visiting behavior transfers from the four training metropolitan areas to the target area Kinki.
    The rule is designed from training-area EDA and applied to Kinki with no target labels, as described in Sections II-C and IV-C.

how reviews work

0 comments
Cite this review

Pith. "Pith review of An Embarrassingly Simple Rule-based Visiting Circulation Approach to Trip Destination Prediction." pith.science (2026). https://pith.science/paper/B7XYHOWJ

@misc{pith2026260725751,
  author       = {Pith},
  title        = {Pith review of: An Embarrassingly Simple Rule-based Visiting Circulation Approach to Trip Destination Prediction},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/B7XYHOWJ}},
  note         = {Machine review of arXiv:2607.25751}
}
read the original abstract

In this paper, we propose the Rule-based Visiting Circulation (RVC) model in tackling the challenge in the IEEE Big Data Cup 2022: Trip Destination Prediction. Given trips containing travel information, personal attributes, origin zones, and their features in the training metropolitan areas, the task is to predict the destination of every trip in a targeted metropolitan area whose destinations are not given at all at the training stage. We highlight the challenges in this destination prediction task -- having no knowledge of the destinations in the targeted metropolitan area. We provide insights from the datasets, in which revisiting behaviors and the relationships between origins and destinations play a crucial role in individuals' trips. Hence, we design a simple but comprehensive method, rule-based visiting circulation, which directly utilizes the origin information and individuals' trip behaviors to determine the destinations in the targeted metropolitan area, i.e., requiring no learning from the four training areas. Experimental results on both offline evaluation and leaderboard submission consistently exhibit the proposed RVC can significantly outperform supervised learning methods and other heuristics. The RVC method eventually brings us to second place in the competition leaderboard.

Figures

Figures reproduced from arXiv: 2607.25751 by the authors.

Figure 1
Figure 1. Occupation and Gender Histogram [PITH_FULL_IMAGE:figures/full_fig_p002_1.png] view at source ↗
Figure 2
Figure 2. Trip Type and Gender Histogram • Gender and Occupation ( [PITH_FULL_IMAGE:figures/full_fig_p002_2.png] view at source ↗
Figure 4
Figure 4. Accuracy of “Destination = Origin” in different [PITH_FULL_IMAGE:figures/full_fig_p003_4.png] view at source ↗
Figures from the paper (2 more)
Figure 5
Figure 5. Figure 5: Elaboration on rule-based visiting circulation by as [PITH_FULL_IMAGE:figures/full_fig_p004_5.png]
Figure 6
Figure 6. Figure 6: The final result of the proposed RVC method (Team “NCKU H.D.D.J.”) in the competition leaderboard. [PITH_FULL_IMAGE:figures/full_fig_p007_6.png]

Discussion (0). Sign in to comment.

Reference graph

Works this paper leans on

17 extracted references · 1 linked inside Pith

  1. [1]

    Random forests.Machine learning, 45(1):5–32, 2001

    Leo Breiman. Random forests.Machine learning, 45(1):5–32, 2001

    2001

  2. [2]

    Similarity encod- ing for learning with dirty categorical variables.Machine Learning, 107(8):1477–1494, 2018

    Patricio Cerda, Ga ¨el Varoquaux, and Bal ´azs K ´egl. Similarity encod- ing for learning with dirty categorical variables.Machine Learning, 107(8):1477–1494, 2018

    2018

  3. [3]

    Amiri, and B

    Liangzhe Chen, Sorour E. Amiri, and B. Aditya Prakash. Automatic segmentation of data sequences. InProceedings of the AAAI Conference on Artificial Intelligence (AAAI), 2018

    2018

  4. [4]

    Will you come back / check-in again? understanding characteristics leading to urban revisitation and re-check- in.Proc

    Zhilong Chen, Hancheng Cao, Huangdong Wang, Fengli Xu, Vassilis Kostakos, and Yong Li. Will you come back / check-in again? understanding characteristics leading to urban revisitation and re-check- in.Proc. ACM Interact. Mob. Wearable Ubiquitous Technol., 4(3), sep 2020

    2020

  5. [5]

    Myers, and Jure Leskovec

    Eunjoon Cho, Seth A. Myers, and Jure Leskovec. Friendship and mobil- ity: User movement in location-based social networks. InProceedings of the 17th ACM SIGKDD International Conference on Knowledge Discovery and Data Mining, KDD ’11, pages 1082–1090, 2011

    2011

  6. [6]

    Measuring and recommending time-sensitive routes from location-based data.ACM Trans

    Hsun-Ping Hsieh, Cheng-Te Li, and Shou-De Lin. Measuring and recommending time-sensitive routes from location-based data.ACM Trans. Intell. Syst. Technol., 5(3), jul 2014

    2014

  7. [7]

    Pseudo-pflow: Development of nationwide synthetic open dataset for people movement based on limited travel survey and open statistical data.arXiv preprint arXiv:2205.00657, 2022

    Takehiro Kashiyama, Yanbo Pang, Yoshihide Sekimoto, and Takahiro Yabe. Pseudo-pflow: Development of nationwide synthetic open dataset for people movement based on limited travel survey and open statistical data.arXiv preprint arXiv:2205.00657, 2022

    arXiv 2022

  8. [8]

    Lightgbm: A highly efficient gradient boosting decision tree

    Guolin Ke, Qi Meng, Thomas Finley, Taifeng Wang, Wei Chen, Weidong Ma, Qiwei Ye, and Tie-Yan Liu. Lightgbm: A highly efficient gradient boosting decision tree. InAdvances in Neural Information Processing Systems, 2017

    2017

Show all 17 references
  1. [9]

    On route planning by inferring visiting time, modeling user preferences, and mining representative trip patterns.Knowledge and Information Systems, 56(3):581–611, 2018

    Cheng-Te Li, Hsin-Yu Chen, Ren-Hao Chen, and Hsun-Ping Hsieh. On route planning by inferring visiting time, modeling user preferences, and mining representative trip patterns.Knowledge and Information Systems, 56(3):581–611, 2018

    2018

  2. [10]

    Hierarchical multi-task graph recurrent network for next poi recommendation

    Nicholas Lim, Bryan Hooi, See-Kiong Ng, Yong Liang Goh, Renrong Weng, and Rui Tan. Hierarchical multi-task graph recurrent network for next poi recommendation. InProceedings of the 45th International ACM SIGIR Conference on Research and Development in Information Retrieval, SI...

    2022

  3. [11]

    Origin-aware next destination recommendation with personalized preference attention

    Nicholas Lim, Bryan Hooi, See-Kiong Ng, Xueou Wang, Yong Liang Goh, Renrong Weng, and Rui Tan. Origin-aware next destination recommendation with personalized preference attention. InProceedings of the 14th ACM International Conference on Web Search and Data Mining, WSDM ’21, p...

    2021

  4. [12]

    Predicting the next location: A recurrent model with spatial and temporal contexts

    Qiang Liu, Shu Wu, Liang Wang, and Tieniu Tan. Predicting the next location: A recurrent model with spatial and temporal contexts. In Proceedings of the Thirtieth AAAI Conference on Artificial Intelligence, AAAI’16, pages 194–200, 2016

    2016

  5. [13]

    Stan: Spatio-temporal attention network for next location recommendation

    Yingtao Luo, Qiang Liu, and Zhaocheng Liu. Stan: Spatio-temporal attention network for next location recommendation. InProceedings of the Web Conference 2021, WWW ’21, pages 2177–2185, 2021

    2021

  6. [14]

    Graph-flashback network for next location recommendation

    Xuan Rao, Lisi Chen, Yong Liu, Shuo Shang, Bin Yao, and Peng Han. Graph-flashback network for next location recommendation. In Proceedings of the 28th ACM SIGKDD Conference on Knowledge Discovery and Data Mining, KDD ’22, pages 1463–1471, 2022

    2022

  7. [15]

    Pflow: Reconstructing people flow recycling large-scale social survey data.IEEE Pervasive Computing, 10(4):27–35, 2011

    Yoshihide Sekimoto, Ryosuke Shibasaki, Hiroshi Kanasugi, Tomotaka Usui, and Yasunobu Shimazaki. Pflow: Reconstructing people flow recycling large-scale social survey data.IEEE Pervasive Computing, 10(4):27–35, 2011

    2011

  8. [16]

    Dingqi Yang, Benjamin Fankhauser, Paolo Rosso, and Philippe Cudre- Mauroux. Location prediction over sparse user mobility traces using rnns: Flashback in hidden states! InProceedings of the Twenty-Ninth International Joint Conference on Artificial Intelligence, IJCAI-20, pages...

    2020

  9. [17]

    Spatio-temporal analysis of passenger travel patterns in massive smart card data.IEEE Transactions on Intelligent Transportation Systems, 18(11):3135–3146, 2017

    Juanjuan Zhao, Qiang Qu, Fan Zhang, Chengzhong Xu, and Siyuan Liu. Spatio-temporal analysis of passenger travel patterns in massive smart card data.IEEE Transactions on Intelligent Transportation Systems, 18(11):3135–3146, 2017

    2017

Pith tools

Reviewed August 1, 2026 · model on record in the stance chip above.