NAB – NYC Taxi Demand (Real Traffic IoT)
- Rows
- 10,320
- Columns
- 2
- Data range
- 2014-07-01 to 2015-01-31T23:30:00
- Auto-refresh
- Re-checks its source every 30 days
Real-world time-series of NYC taxi passenger counts aggregated every 30 minutes. IoT/sensor-adjacent streaming data with labeled anomalies corresponding to NYC events. Classic NAB benchmark dataset.
AI analysis
NYC Taxi Demand Dataset — Analytical Overview
1. Dataset Representation & Correlation Value This dataset captures real-world NYC taxi passenger counts aggregated at 30-minute intervals, sourced directly from the Numenta Anomaly Benchmark (NAB) repository at github.com/numenta/NAB — a well-maintained, academically recognized benchmark library for time-series anomaly detection. With 10,320 rows spanning a continuous temporal sequence, the dataset represents roughly 215 days of uninterrupted 30-minute readings, making it exceptionally well-suited for time-series correlation studies. Its embedded anomaly labels tied to real NYC events (holidays, weather incidents, public gatherings) elevate it beyond a simple traffic log into a labeled event-response dataset, offering strong potential for correlating urban mobility patterns against external signals such as weather, event calendars, or transit disruptions.
2. Data Quality Observations Data quality here is notably strong. Zero null cells are reported across both columns (0 of 20,640 total cells), and the timestamp column exhibits 10,320 distinct values matching the row count exactly — confirming no duplicate timestamps and a perfectly regular time index with no gaps detectable at the index level. The duplicate-row count is pending final Parquet-phase recomputation, but the timestamp uniqueness effectively rules out meaningful row-level duplication. The only minor caveat is that sample row preview is unavailable pending pipeline wire-up, so field-level formatting assumptions (e.g., timestamp timezone, exact ISO format) should be validated before joining. Overall, this is among the cleaner IoT time-series datasets one would encounter in practice.
3. Key Column Distributions The value column (integer passenger counts per 30-minute window) tells a rich story through its statistics. The mean of 15,137.6 sits noticeably below the median of 16,778, and combined with a negative skew of -0.452, this indicates a left-skewed distribution — the bulk of readings cluster in moderate-to-high demand territory, but a meaningful lower tail pulls the mean down, likely reflecting overnight low-demand periods. The IQR spans 10,262 (Q1) to 19,839 (Q3), a wide band of ~9,577 passengers suggesting substantial daily cyclicality. The range is dramatic — min=8 to max=39,197 — with only 2 flagged outliers, which is impressively low for IoT sensor data and suggests the NAB curation process has already handled gross anomalies. Those 2 outliers likely correspond to documented NYC event spikes and deserve preservation rather than removal for any anomaly correlation study.
4. Recommended Join Key Columns The timestamp column is the definitive join key, boasting zero nulls and 10,320 distinct values with full cardinality — a 1:1 record-to-timestamp ratio. For cross-dataset joins, this column should be normalized to UTC and truncated to 30-minute intervals to align with partner datasets that may use hourly or daily granularity (requiring aggregation). A derived date column and a hourofday / dayofweek column extracted from the timestamp would significantly expand joinability to datasets with coarser temporal resolution, enabling both exact-timestamp joins and time-bucket fuzzy joins.
5. Recommended Pairing Datasets for Correlation Discovery Several dataset categories would pair powerfully with this data:
- NYC Weather / NOAA Station Data — Temperature, precipitation, and wind readings at 30-minute or hourly resolution could reveal how weather suppresses or spikes taxi demand, with the negative-skew low-demand tail potentially attributable to severe weather events. - NYC MTA Subway Ridership — Correlated or inversely correlated transit ridership (e.g., subway outages driving taxi demand spikes) would directly test modal substitution hypotheses. - NYC Public Events Calendar — Concerts, sports games, and parades likely explain the documented NAB anomaly labels; joining event timestamps could quantify demand lift per event category. - Uber/Lyft Rideshare Volume Data — If available at comparable granularity, rideshare data could reveal competitive demand splitting and combined mobility trends. - NYC Holiday & School Calendar — Given the visible cyclicality implied by the wide IQR, holiday and academic calendars would help decompose structural demand patterns from anomalous ones.
Columns
- timestamp (date)
- value (integer)