{"id":"e31bcbe9-cbe6-4b16-a7af-12344c397b12","arxiv_id":"2505.03816","paper_version":1,"verdict":"CONDITIONAL","confidence":"MODERATE","novelty_score":2.0,"correctness_risk":"medium","formal_verification":"none","parameter_count":3,"one_line_summary":"Using one month of NYC taxi data and a Dhaka food delivery dataset, the authors report that Manhattan has the most taxi pickups, Staten Island the fewest, and Pathao orders peak on Fridays and Saturdays.","lead":"This paper analyzes taxi trips in New York City and food delivery orders in Dhaka, Bangladesh, to map when and where demand peaks. It shows how routine data analysis, including forecasting and clustering, can help operators plan fleets and staffing, though the evidence is limited to one month of NYC data and the Dhaka dataset has no location or timestamp information.","discovery_kind":"extension","skeptic_critique":{"model":"deepseek-v4-flash","headline":"The weekly-seasonality claim rests on ~31 daily January points, a test window shorter than one seasonal cycle, and an in-sample RMSE used as the headline evidence.","rationale":"The reader and I converge on the same load-bearing weakness, so I cannot identify a more central problem. The strongest claim is a forecasting claim, and everything needed to support it is the validation of the fitted SARIMAX. That validation currently depends on one month of daily data, an 80:20 split giving a test set shorter than the seasonal period, and a headline RMSE that is the in-sample number. None of these are fatal to the descriptive parts of the paper, and the authors do flag the data limitation in Section 7, but the operational conclusion in Section 5.3 is not warranted by the evidence as written. A full-year rolling evaluation against a seasonal-naive baseline is the direct way to settle whether the concern lands; the raw data for 2013 is already available and referenced. I would not change the conditional verdict: the paper can be accepted if this validation is supplied and the claims are tempered accordingly. The paper deserves credit for clearly described EDA and for an explicit acknowledgment of the single-month limitation, but the central forecasting claim needs the proposed robustness check before it can be taken as evidence of practical forecasting skill.","tokens_in":7894,"tokens_out":4680,"duration_ms":49094,"concrete_test":"Obtain daily pickup counts for the full 2013 NYC TLC dataset, which the paper states is available. Fit the same SARIMAX(1,0,1)(1,0,1,7) on a rolling 84-day training window, forecast the next 14 days, and repeat monthly across the year. Compare RMSE and MAE against (i) a seasonal-naive model that predicts each day using the average of the same weekday in the training window and (ii) a nonseasonal ARIMA(1,0,1). Report estimated seasonal AR and MA coefficients with standard errors and residual autocorrelation at lag 7. If the SARIMAX does not consistently beat the seasonal-naive baseline, or if the seasonal coefficients are insignificant, the Section 5.3 claim of effectively captured weekly seasonality is not supported.","verdict_should_be":"UNCHANGED","load_bearing_attack":"The central claim in Section 5.3 is that a SARIMAX model with order=(1,0,1) and seasonal_order=(1,0,1,7) effectively captured weekly seasonality, with RMSE 734.944. The supporting evidence is under-powered in three linked ways. First, Section 3.1.1 restricts the NYC data to January 2013, about 31 daily observations; with the 80:20 split of Section 3.3, this leaves roughly 6 test points, less than one full weekly cycle. An out-of-sample RMSE computed over fewer than one seasonal period cannot validate a period-7 seasonal model. Second, the cited RMSE of 734.944 is the in-sample value; the out-of-sample value of 204.1525 is reported but not used for the conclusion, and no forecast horizon, baseline, or uncertainty interval is provided. Third, fitting seasonal AR and MA terms at lag 7 from about four weekly cycles is statistically fragile, so an apparent weekly pattern can reflect overfitting rather than a stable seasonal process. The authors themselves note in Section 7 that the single-month scope may limit broader applicability. The descriptive EDA is plausible and consistent with the figures, but the operational forecasting claim in Section 5.3 is not established by the evidence as presented.","agreement_with_reader":"agree"},"referee_report":{"model":"deepseek-v4-flash","summary":"The paper analyzes two urban transportation datasets: NYC Taxi Trip data from January 2013 and Pathao Food Delivery data from Dhaka. It performs exploratory data analysis, borough-level geospatial summaries, K-means clustering of pickup/dropoff coordinates, and a SARIMAX time-series forecast for taxi demand. The descriptive findings are that Manhattan has the highest taxi pickup volume, Staten Island the lowest, taxi trips tend to be longer later in the day and on weekends, and Pathao orders peak on Fridays and Saturdays. The paper also claims that the SARIMAX model effectively captured weekly seasonality with an RMSE of 734.944 and that this can support fleet management and driver scheduling.","tokens_in":8159,"tokens_out":4969,"duration_ms":52203,"significance":"If the forecasting claim were supported, the work would be practically relevant for taxi and food-delivery operators. The descriptive parts are straightforward and generally consistent with the displayed figures, and the authors are honest in Section 7 that the single-month NYC sample limits generalizability. The strengths are the use of real-world data, the clear exploratory summaries, and the explicit acknowledgment of scope limitations. However, the time-series evidence is underpowered: a period-7 seasonal model is fitted to roughly 31 daily observations with a test window shorter than one seasonal cycle, and the headline RMSE is the in-sample value. With additional baselines, proper out-of-sample evaluation, and more cautious claims, this could become a useful applied case study, but the central forecasting claim is not currently established.","major_comments":[{"comment":"The central forecasting claim is not supported by the reported experimental setup. The NYC data are restricted to January 2013, about 31 daily observations, and the 80:20 train-test split leaves roughly 6 test points, fewer than one full weekly cycle. A SARIMAX model with seasonal order (1,0,1,7) cannot be validated on a test window shorter than the seasonal period, and estimating seasonal AR and MA terms from roughly 25 training points is statistically fragile. The authors should either use a longer series, validate with multiple seasonal cycles, or explicitly present the single-month result as a methodological illustration rather than as evidence that weekly seasonality has been captured.","section":"§3.1.1, §3.3, §5.3"},{"comment":"The RMSE reporting does not justify the conclusion drawn. The quoted RMSE of 734.944 is the in-sample value; the out-of-sample value of 204.1525 is reported but not interpreted, and no baseline model, forecast horizon, or confidence interval is provided. In-sample RMSE cannot serve as evidence of forecast skill, and the out-of-sample value needs context such as a seasonal-naive baseline or a non-seasonal ARIMA comparison. Section 5.3 should be rewritten so that the conclusion is drawn only from out-of-sample performance with suitable baselines.","section":"§5.3"},{"comment":"The clustering analysis does not support the claim that K-means identifies high- and low-demand regions. The number of clusters K=15 is chosen without a selection criterion or stability check, and the clustering input is raw latitude and longitude coordinates, not demand counts. The resulting figure likely shows geographic groupings rather than demand hotspots. To support the claim, the authors should combine cluster assignments with trip counts and justify the choice of K or test its robustness.","section":"§4.4, §5.4"},{"comment":"The comparative framing overstates what is actually analyzed. Section 3.1.2 states that the Pathao dataset has no geospatial coordinates and no datetime attribute, so geospatial analysis and time-series forecasting are performed only for the NYC taxi data. The Pathao contribution is limited to exploratory analysis of day-of-week and hour-of-day patterns. The introduction and abstract should be reframed to make this asymmetry explicit, rather than implying that both datasets support the full geospatial and temporal pipeline.","section":"§3.1.2, §5, Abstract"}],"minor_comments":[{"comment":"The title contains an apparent typo: 'A S TUDY' should be 'A STUDY'.","section":"Title"},{"comment":"Section 3.3 says an 80:20 train-test split was performed on both datasets, but no model is evaluated on the Pathao data; clarify that the split applies only to the time-series analysis of the NYC taxi data.","section":"§3.3"},{"comment":"The text says the data are de-seasonalized, while Figure 7 is labeled 'De-Trended Time Series'; make the terminology consistent.","section":"§4.3 and Figure 7"},{"comment":"The phrase 'late hours (15th and 23rd hour)' is ambiguous; it should specify 15:00 and 23:00, and the claims about hourly peaks should be tied directly to the corresponding figures.","section":"§5.1"},{"comment":"Reference [9] appears in the reference list but is not cited in the body of the paper, and several references lack complete venue or pagination details; the bibliography needs cleanup.","section":"References"},{"comment":"The paper does not include a data availability or reproducibility statement, and no code is provided; adding these would strengthen the contribution.","section":"General"}],"recommendation":"major_revision","confidential_remarks":"The manuscript is best viewed as an applied descriptive case study. The novelty is modest, and the SARIMAX section requires substantial revision before the forecasting claim can be taken seriously. If the authors are able to extend the analysis to a longer time period and provide baseline comparisons, the paper could be suitable for an applied venue; in its current form, the central claim is not established by the evidence presented."},"author_rebuttal":null,"desk_editor":{"model":"deepseek-v4-flash","letter":"Punchline: this is a workmanlike EDA on two public datasets, but the forecasting claim is not supported by the evidence. The descriptive patterns are probably correct; the SARIMAX conclusion is not.\n\nWhat the paper does well: it gives a clear, honest account of the Pathao Food dataset, which has been analyzed less often, and it openly admits in Section 7 that the one-month NYC Taxi window limits the scope. The bar charts and cluster maps appear consistent with the underlying data. For a reader who wants a straightforward template for taxi-demand EDA, this is not a bad starting point.\n\nSoft spots: the load-bearing claim in Section 5.3 is that a (1,0,1)(1,0,1,7) SARIMAX 'effectively captured weekly seasonality' with an in-sample RMSE of 734.94. The dataset is January 2013 only, roughly 31 daily counts. The 80/20 split leaves six test points, less than one seasonal cycle. Out-of-sample RMSE (204.15) is reported but unused as evidence, and there is no baseline, no forecast horizon, no uncertainty interval. You cannot validate a period-7 seasonal process on fewer than seven test observations. The Pathao data also lack geospatial and datetime attributes, so the paper's own frame of a 'geospatial and temporal' comparison between NYC and Dhaka cannot hold; the Pathao analysis is EDA only. The K-means step uses an unexplained K=15. No code or data are provided, so nothing is reproducible.\n\nVerdict: this is a student-scale case study, not a contribution to the literature. The NYC taxi findings replicate known patterns. It would be a reasonable teaching illustration but not a publishable research claim as written. I would not send it to a serious referee in this form. If the authors tone down the forecasting language, add baselines and error bars, publish code, and reframe the Pathao part as descriptive EDA, it might become a workshop-level paper. For now, desk reject seems right.","headline":"A modest EDA case study whose forecast claim outruns its data; the descriptive parts are fine, the SARIMAX evidence is not.","tokens_in":8690,"tokens_out":3443,"would_cite":false,"duration_ms":32202,"reading_group":"no","serious_thinker":"yes","would_accept_peer_review":false},"rs_alignment":null,"lean_confirmation":null,"pith_extraction":{"msc":[],"pacs":[],"model":"deepseek-v4-flash","headline":"A weekly-season SARIMAX model fit to one month of NYC taxi pickups is claimed to forecast daily demand well enough to guide fleet scheduling.","keywords":["urban transportation","taxi demand forecasting","SARIMAX","NYC taxi data","food delivery demand","weekly seasonality","geospatial clustering","K-means"],"falsifier":"Fit the same SARIMAX(1,0,1)(1,0,1,7) model to several months of NYC taxi data with a holdout of at least four weeks, and compare its out-of-sample RMSE with a plain ARIMA(1,0,1) that has no seasonal term; if the seasonal model is not better, or if significant autocorrelation remains at lag 7, the weekly-seasonality claim would not survive.","tokens_in":7686,"feed_emoji":"🚕","tokens_out":8774,"duration_ms":78898,"temperature":0.7,"pith_summary":"This paper tries to establish that demand in two urban services—yellow taxis in New York City and food deliveries in Dhaka—is regular enough to summarize, map, and forecast. Its central quantitative claim is that a SARIMAX model with a weekly seasonal term, fit to daily NYC taxi pickup counts from January 2013, predicts daily demand well enough to guide fleet placement and driver scheduling, with an in-sample RMSE of 734.944 and an out-of-sample RMSE of 204.1525. The paper also claims that Manhattan leads pickup demand, Queens has the longest average trips, and Pathao food orders peak on Fridays and Saturdays. A reader should care because the same pipeline from exploratory analysis to seasonal forecasting could transfer to other cities and delivery fleets.","feed_headline":"Weekly cycle forecasts NYC taxi demand from one month of data","feed_subtitle":"SARIMAX with a 7-day seasonal term reports RMSE 735 in sample and 204 out of sample, enough for fleet scheduling.","key_machinery":"The central object is the SARIMAX model—seasonal autoregressive integrated moving average with exogenous variables—configured with order (1,0,1) and seasonal order (1,0,1,7). The seasonal period of 7 is the mechanism that encodes the weekly cycle in daily pickup counts, and the paper's case rests on that term's ability to reduce forecast error to the reported RMSE values. Supporting machinery includes K-means clustering with 15 clusters on pickup and dropoff coordinates, which identifies geographic high- and low-demand zones, and feature engineering that extracts hour, day, day of week, and trip duration from datetime fields.","core_discovery":"Stated on the paper's own terms, the central discovery is that daily New York City taxi demand in January 2013 carries a weekly seasonal structure that a SARIMAX(1,0,1)(1,0,1,7) model captures: the fitted model yields an in-sample RMSE of 734.944 and an out-of-sample RMSE of 204.1525 on a short test window. The authors take these numbers as evidence that the model 'effectively captured the weekly seasonality' and that it can help operators forecast demand patterns for optimized fleet management and efficient driver scheduling. The companion descriptive findings are that Manhattan dominates pickup volume, Staten Island has the least activity, Queens shows the longest average trip distance and duration, trips lengthen later in the day and on weekends, and Pathao food delivery demand peaks on Friday and Saturday.","pith_inferences":["A testable extension the paper does not report is a head-to-head comparison with a non-seasonal ARIMA on the same holdout, which would isolate the seasonal term's marginal contribution.","Aggregating the Pathao records by day of week and hour would test whether the Friday-Saturday ordering peak persists after accounting for how many restaurants are open.","Repeating the fit on other months of 2013 would show whether the weekly seasonal order and the Manhattan/Queens/Staten Island rankings are stable, which the one-month design does not establish."],"forward_implications":["Taxi operators who adopt the model can shift drivers toward late-day and weekend hours where pickups and trip lengths rise, and concentrate vehicles in Manhattan while reducing idle capacity in Staten Island.","Retraining the same SARIMAX configuration on longer NYC taxi histories should produce rolling daily demand forecasts suitable for dispatch planning, if the weekly pattern holds across months.","Food delivery operators in Dhaka can act on the weekday/weekend demand split by scheduling more riders and preparing popular items on Fridays and Saturdays, even though the Pathao dataset lacks timestamps and coordinates for a full time-series model.","Positioning idle taxis near the high-demand clusters identified by K-means should cut wait times, provided those clusters remain stable over the weeks the model covers."],"supporting_citations":[{"why":"Supplies the ARIMA forecasting baseline that the paper's SARIMAX extends to seasonal transport demand.","marker":"[5]"},{"why":"Shows ARIMA outperforming LSTM on a forecasting task, motivating the choice of the classical seasonal ARIMA family.","marker":"[6]"},{"why":"Compares ARIMA, SARIMA, and LSTM and demonstrates SARIMA-type models capturing seasonal patterns, supporting the seasonal order used here.","marker":"[7]"},{"why":"Provides a direct precedent of a SARIMA model with a seasonal order for forecasting transport passenger demand.","marker":"[8]"},{"why":"Grounds the NYC taxi dataset as a source for spatiotemporal demand and community structure analysis.","marker":"[3]"},{"why":"Prior exploratory analysis of NYC taxi fare data that identified high-demand boroughs and informs the descriptive comparisons.","marker":"[4]"}],"fun_headline_variants":["One month of NYC taxi data reveals weekly demand cycle","SARIMAX captures weekly cycle from 31 days of NYC taxis","Short data wins: NYC taxi demand forecast from single month","NYC taxi weekly pattern spotted with just one month of data"],"cache_read_input_tokens":3200,"weakest_assumption_plain":"The central claim depends on the assumption that one month of daily trip counts—January 2013—contains enough repeated weekly cycles to estimate and test a weekly seasonal model, with a test period of only about six days.","fun_headline_variants_meta":{"raw":{"variants":["One month of NYC taxi data reveals weekly demand cycle","SARIMAX captures weekly cycle from 31 days of NYC taxis","Short data wins: NYC taxi demand forecast from single month","NYC taxi weekly pattern spotted with just one month of data"]},"model":"deepseek-v4-flash","effort":"low","cost_usd":0.000648,"raw_usage":{"total_tokens":2945,"prompt_tokens":888,"completion_tokens":2057,"prompt_tokens_details":{"cached_tokens":384},"prompt_cache_hit_tokens":384,"prompt_cache_miss_tokens":504,"completion_tokens_details":{"reasoning_tokens":1987}},"tokens_in":504,"tokens_out":2057,"duration_ms":14930,"temperature":1.0,"reasoning_tokens":1987,"cache_read_input_tokens":384,"cache_creation_input_tokens":0},"cache_creation_input_tokens":0},"created_at":"2026-08-16T04:18:02.536595+00:00","model_set":{"reader":"deepseek-v4-flash"},"falsifier":"Fit the same SARIMAX(1,0,1)(1,0,1,7) model to several months of NYC taxi data with a holdout of at least four weeks, and compare its out-of-sample RMSE with a plain ARIMA(1,0,1) that has no seasonal term; if the seasonal model is not better, or if significant autocorrelation remains at lag 7, the weekly-seasonality claim would not survive.","supporting_citations":[{"cited_title":"ARIMA model for accurate time series stocks forecasting","cited_arxiv_id":null,"evidence_quote":"Supplies the ARIMA forecasting baseline that the paper's SARIMAX extends to seasonal transport demand."},{"cited_title":"Time Series Forecasting using LSTM and ARIMA","cited_arxiv_id":null,"evidence_quote":"Shows ARIMA outperforming LSTM on a forecasting task, motivating the choice of the classical seasonal ARIMA family."},{"cited_title":"Forecasting Surabaya–Jakarta Train Passengers with SARIMA model","cited_arxiv_id":null,"evidence_quote":"Provides a direct precedent of a SARIMA model with a seasonal order for forecasting transport passenger demand."},{"cited_title":"Revealing spatiotemporal travel demand and community structure characteristics with taxi trip data: A case study of New York City","cited_arxiv_id":null,"evidence_quote":"Grounds the NYC taxi dataset as a source for spatiotemporal demand and community structure analysis."},{"cited_title":"Data Exploratory on Taxi Data in New York City","cited_arxiv_id":null,"evidence_quote":"Prior exploratory analysis of NYC taxi fare data that identified high-demand boroughs and informs the descriptive comparisons."}],"review_version":1}