Datasets:

Modalities:
Text
ArXiv:
License:
Dataset Viewer
Auto-converted to Parquet Duplicate
agent_id
int64
149
500k
anomaly_type
stringclasses
10 values
anomaly_name
stringclasses
10 values
subtype
stringclasses
18 values
test_start_time
stringdate
2024-09-16 06:50:00
2024-10-13 11:55:00
149
2b
role_violation
2b_student_worker_routine
2024-09-16 09:05:00
797
1d
temporal_shift
1d_temporal_shift
2024-09-16 08:05:00
807
1a
replace_activity
1a_replace_activity
2024-09-16 08:05:00
1,483
4c
periodic_anomaly
4c_periodic_replacement
2024-09-16 08:05:00
1,621
4c
periodic_anomaly
4c_periodic_replacement
2024-09-16 08:05:00
1,642
1d
temporal_shift
1d_temporal_shift
2024-09-16 08:05:00
1,649
4c
periodic_anomaly
4c_periodic_recurring_visit
2024-09-16 09:00:00
1,985
1c
insert_activity
1c_insert_activity
2024-09-16 10:40:00
2,197
1a
replace_activity
1a_replace_activity
2024-09-18 10:45:00
2,205
2b
role_violation
2b_student_worker_routine
2024-09-16 09:00:00
2,269
1b
delete_activity
1b_delete_activity
2024-09-16 10:35:00
2,340
2b
role_violation
2b_student_worker_routine
2024-09-16 09:10:00
2,358
4c
periodic_anomaly
4c_periodic_deletion
2024-09-16 08:30:00
2,415
4c
periodic_anomaly
4c_periodic_deletion
2024-09-16 10:40:00
2,447
1c
insert_activity
1c_insert_activity
2024-09-16 09:05:00
2,516
2b
role_violation
2b_student_work_commute
2024-09-16 09:05:00
2,609
2b
role_violation
2b_student_worker_routine
2024-09-16 09:05:00
2,972
1f
mode_anomaly
1f_mode_anomaly
2024-09-17 09:45:00
3,057
1a
replace_activity
1a_replace_activity
2024-09-17 07:55:00
3,166
1d
temporal_shift
1d_temporal_shift
2024-09-16 08:05:00
3,168
4c
periodic_anomaly
4c_periodic_deletion
2024-09-16 09:00:00
3,176
1b
delete_activity
1b_delete_activity
2024-09-16 08:05:00
3,491
2b
role_violation
2b_student_worker_routine
2024-09-16 09:05:00
3,492
2b
role_violation
2b_student_worker_routine
2024-09-16 09:05:00
3,544
4c
periodic_anomaly
4c_periodic_recurring_visit
2024-09-16 09:10:00
3,632
1c
insert_activity
1c_insert_activity
2024-09-16 08:05:00
3,694
4c
periodic_anomaly
4c_periodic_deletion
2024-09-16 08:30:00
3,812
1c
insert_activity
1c_insert_activity
2024-09-16 08:00:00
3,826
1e
duration_anomaly
1e_duration_anomaly
2024-09-16 09:15:00
3,840
1d
temporal_shift
1d_temporal_shift
2024-09-16 08:05:00
3,914
5a
geofence_violation
5a_geofence_violation
2024-09-16 09:00:00
4,009
1e
duration_anomaly
1e_duration_anomaly
2024-09-16 08:05:00
4,063
4c
periodic_anomaly
4c_periodic_deletion
2024-09-16 09:05:00
4,168
2b
role_violation
2b_worker_school_routine
2024-09-16 08:05:00
4,305
1b
delete_activity
1b_delete_activity
2024-09-16 08:05:00
4,391
5a
geofence_violation
5a_geofence_violation
2024-09-16 07:50:00
4,513
4c
periodic_anomaly
4c_periodic_recurring_visit
2024-09-16 10:20:00
4,514
1c
insert_activity
1c_insert_activity
2024-09-16 08:05:00
4,643
1a
replace_activity
1a_replace_activity
2024-09-16 08:05:00
4,738
1a
replace_activity
1a_replace_activity
2024-09-18 08:50:00
4,792
4c
periodic_anomaly
4c_periodic_deletion
2024-09-16 08:30:00
4,953
2b
role_violation
2b_worker_errand_during_workday
2024-09-16 08:05:00
4,979
2b
role_violation
2b_worker_school_routine
2024-09-16 10:35:00
5,000
1f
mode_anomaly
1f_mode_anomaly
2024-09-16 08:05:00
5,019
4c
periodic_anomaly
4c_periodic_replacement
2024-09-16 08:05:00
5,027
1a
replace_activity
1a_replace_activity
2024-09-16 08:05:00
5,117
2b
role_violation
2b_worker_errand_during_workday
2024-09-16 08:05:00
5,167
1e
duration_anomaly
1e_duration_anomaly
2024-09-16 10:35:00
5,484
1d
temporal_shift
1d_temporal_shift
2024-09-16 08:05:00
5,519
2b
role_violation
2b_student_work_commute
2024-09-16 09:00:00
5,521
1e
duration_anomaly
1e_duration_anomaly
2024-09-16 09:10:00
5,581
4c
periodic_anomaly
4c_periodic_recurring_visit
2024-09-17 08:55:00
5,629
4c
periodic_anomaly
4c_periodic_replacement
2024-09-16 08:05:00
5,712
4c
periodic_anomaly
4c_periodic_deletion
2024-09-16 09:40:00
5,720
1a
replace_activity
1a_replace_activity
2024-09-16 10:40:00
5,822
1a
replace_activity
1a_replace_activity
2024-09-16 08:05:00
5,843
1a
replace_activity
1a_replace_activity
2024-09-20 10:35:00
6,026
4c
periodic_anomaly
4c_periodic_recurring_visit
2024-09-16 09:05:00
6,050
1a
replace_activity
1a_replace_activity
2024-09-16 09:00:00
6,314
1f
mode_anomaly
1f_mode_anomaly
2024-09-17 08:35:00
6,380
2b
role_violation
2b_student_work_commute
2024-09-16 09:00:00
6,506
2b
role_violation
2b_student_work_commute
2024-09-16 09:10:00
6,556
1d
temporal_shift
1d_temporal_shift
2024-09-16 08:05:00
6,649
1c
insert_activity
1c_insert_activity
2024-09-16 08:05:00
6,661
5a
geofence_violation
5a_geofence_violation
2024-09-16 08:05:00
6,753
4c
periodic_anomaly
4c_periodic_recurring_visit
2024-09-16 08:05:00
6,771
1d
temporal_shift
1d_temporal_shift
2024-09-16 08:05:00
6,803
1d
temporal_shift
1d_temporal_shift
2024-09-16 08:05:00
6,859
4c
periodic_anomaly
4c_periodic_recurring_visit
2024-09-16 08:05:00
6,874
4c
periodic_anomaly
4c_periodic_recurring_visit
2024-09-16 09:05:00
7,035
1a
replace_activity
1a_replace_activity
2024-09-16 08:05:00
7,080
1e
duration_anomaly
1e_duration_anomaly
2024-09-16 08:05:00
7,114
2b
role_violation
2b_worker_school_routine
2024-09-16 08:05:00
7,287
1e
duration_anomaly
1e_duration_anomaly
2024-09-16 08:05:00
7,327
1f
mode_anomaly
1f_mode_anomaly
2024-09-16 08:05:00
7,390
5a
geofence_violation
5a_geofence_violation
2024-09-16 09:05:00
7,558
2b
role_violation
2b_student_worker_routine
2024-09-16 09:00:00
7,684
2b
role_violation
2b_worker_school_routine
2024-09-16 08:05:00
7,799
4c
periodic_anomaly
4c_periodic_deletion
2024-09-16 09:05:00
7,806
1e
duration_anomaly
1e_duration_anomaly
2024-09-16 08:05:00
7,889
4c
periodic_anomaly
4c_periodic_replacement
2024-09-16 08:05:00
7,898
1d
temporal_shift
1d_temporal_shift
2024-09-16 08:05:00
7,942
1f
mode_anomaly
1f_mode_anomaly
2024-09-16 08:05:00
7,943
1a
replace_activity
1a_replace_activity
2024-09-18 11:45:00
8,160
1c
insert_activity
1c_insert_activity
2024-09-16 09:00:00
8,200
1f
mode_anomaly
1f_mode_anomaly
2024-09-16 08:05:00
8,213
4c
periodic_anomaly
4c_periodic_deletion
2024-09-16 09:10:00
8,574
1b
delete_activity
1b_delete_activity
2024-09-16 08:05:00
8,613
4c
periodic_anomaly
4c_periodic_deletion
2024-09-16 08:05:00
8,708
2b
role_violation
2b_student_worker_routine
2024-09-16 09:05:00
8,767
4c
periodic_anomaly
4c_periodic_replacement
2024-09-16 08:05:00
8,774
1f
mode_anomaly
1f_mode_anomaly
2024-09-16 08:05:00
8,803
1d
temporal_shift
1d_temporal_shift
2024-09-16 08:05:00
8,867
2b
role_violation
2b_student_worker_routine
2024-09-16 09:30:00
8,869
5a
geofence_violation
5a_geofence_violation
2024-09-16 10:35:00
8,891
4c
periodic_anomaly
4c_periodic_deletion
2024-09-16 08:05:00
9,004
2b
role_violation
2b_student_worker_routine
2024-09-16 09:05:00
9,056
4c
periodic_anomaly
4c_periodic_recurring_visit
2024-09-20 08:50:00
9,094
1c
insert_activity
1c_insert_activity
2024-09-16 08:05:00
9,257
2b
role_violation
2b_student_worker_routine
2024-09-16 09:05:00
End of preview. Expand in Data Studio

SF-ANOMALIES Public Release Files

SF-ANOMALIES is a behavioral anomaly benchmark dataset built on SF-LIFE. Each released agent has a clean normal history followed by a test period containing one behavioral or contextual anomaly realized as a route-valid multimodal trajectory.

Licenses and Attribution

SF-ANOMALIES is © 2026 The SF-ANOMALIES authors.

The dataset in the data folder is made available under the Open Data Commons Attribution License (ODC-By): https://opendatacommons.org/licenses/by/ The license text is included in LICENSE_DATA.txt.

The script in the scripts folder is made available under the New BSD License: https://opensource.org/license/bsd-3-clause The license text is included in LICENSE_SCRIPTS.txt

Building locations and road routes derive from OpenStreetMap: map data © OpenStreetMap contributors, available under the Open Database License (ODbL): https://www.openstreetmap.org/copyright

When using the SF-ANOMALIES dataset or scripts, please cite the following paper or its follow-up journal version:

Mauryan Uppalapati and Carola Wenk. SF-ANOMALIES: A Mobility Dataset for Behavioral and Contextual Anomaly Detection, in The 3rd ACM SIGSPATIAL International Workshop on Geospatial Anomaly Detection (GeoAnomalies '26), Riverside, CA, USA, 2026. https://doi.org/10.1145/3849730.3849826

A short paper on the dataset appeared in The 34th ACM International Conference on Advances in Geographic Information Systems (SIGSPATIAL '26), Riverside, CA, USA, 2026. https://doi.org/10.1145/3841645.3843019

Release Layout

README.md
data/
  LICENSE_DATA.txt
  trajectories_1hz/
    agent_<agent_id>.parquet
  windows_1hz/
    agent_<agent_id>.parquet
  test_period_1min.parquet
  labels/
    agent_level_ground_truth.csv
    temporal_level_ground_truth.csv

scripts/
  LICENSE_SCRIPTS.txt
  downsample_trajectories.py

File Descriptions

  • data/trajectories_1hz/: Directory containing 5,000 complete 1 Hz test-period trajectory files, one per released SF-LIFE agent. Files are named agent_<agent_id>.parquet, where <agent_id> is the SF-LIFE agent identifier listed in data/labels/agent_level_ground_truth.csv. Each file holds one row per second from the agent's test_start_time to Oct 21 2024 00:00, with no gaps (14,835,585,300 rows in total).
  • data/windows_1hz/: The same 1 Hz rows restricted to each agent's anomaly windows: the stretches of the test period whose trajectory the anomaly changed, with a margin at each end (7,125 windows in total, 1 to 8 per agent, 552,206,460 rows). One file per agent, named as above.
  • data/test_period_1min.parquet: The test period of all 5,000 agents at one row per minute (247,259,755 rows): the 1 Hz rows at second :00.
  • data/labels/agent_level_ground_truth.csv: One row per released agent.
  • data/labels/temporal_level_ground_truth.csv: One row per semantic anomaly window. Therefore an anomaly that recurs (periodic and some role anomalies) has one row per occurrence for the same agent.
  • scripts/downsample_trajectories.py: Utility for producing lower temporal resolutions from the 1 Hz trajectories.

Trajectory Convention

The source simulation covers 70 days, Aug 12 2024 00:00 to Oct 21 2024 00:00. Anomaly injection is restricted to the evaluation horizon after the nominal 35-day clean-history period. The test period of each agent starts at its test_start_time, an activity boundary of that agent on or after Sep 16 2024 (Sep 16 for 4,299 of the 5,000 agents; Oct 13 11:55 at the latest), and ends at Oct 21 2024 00:00 (exclusive). Every labelled anomaly lies in the test period.

Event-level anomaly intervals are provided in temporal_level_ground_truth.csv; join those windows to trajectory rows by agent_id (the trajectory column agent) and timestamp. The temporal labels mark when the anomalous behavior happens, and the 1 Hz windows mark where the trajectory changed. Every temporal label overlaps one of the agent's 1 Hz windows; 460 of them (duration and mode anomalies) extend beyond it.

Trajectory Schema

Each trajectory parquet file has these columns:

Column Meaning
timestamp Local clock time (America/Los_Angeles), without a time zone. The whole simulation is in daylight time, so UTC = local + 7 hours.
agent Agent identifier (the SF-LIFE agent id).
modality Released timestamp-level modality label (int8): 0 stationary, 1 walking, 2 bike, 3 car, 5 bus, 6 rail.
longitude WGS84 longitude.
latitude WGS84 latitude.

Agent-Level Ground Truth

agent_level_ground_truth.csv has 5,000 rows, one per agent. Key columns:

Column Meaning
agent_id SF-LIFE agent identifier.
anomaly_type, anomaly_name Anomaly taxonomy labels.
subtype Anomaly subtype, e.g. 2a_child_late_night_outing.
test_start_time Start of the agent's test period, local time.

Temporal-Level Ground Truth

temporal_level_ground_truth.csv has 7,231 rows, one per semantic anomaly window. This is the primary event-level label file. For example, an agent with six periodic recurring visits appears in six temporal rows.

Column Meaning
agent_id Agent identifier.
anomaly_type, anomaly_name Anomaly taxonomy labels.
anomaly_start_time, anomaly_end_time Event-level anomaly interval in local schedule time.

Basic Loading Example

import pandas as pd

agents = pd.read_csv("data/labels/agent_level_ground_truth.csv")
windows = pd.read_csv("data/labels/temporal_level_ground_truth.csv")

agent_id = agents.loc[0, "agent_id"]
agent_windows = windows[windows["agent_id"] == agent_id]

traj_1hz = pd.read_parquet(f"data/trajectories_1hz/agent_{agent_id}.parquet")
windows_1hz = pd.read_parquet(f"data/windows_1hz/agent_{agent_id}.parquet")
traj_1min = pd.read_parquet("data/test_period_1min.parquet", filters=[("agent", "==", agent_id)])

Downsampling Example

python scripts/downsample_trajectories.py \
  --input data/trajectories_1hz \
  --output data/trajectories_1min \
  --seconds 60

The script preserves the relative folder layout, so data/trajectories_1hz/agent_123.parquet becomes data/trajectories_1min/agent_123.parquet. At --seconds 60 it gives the same rows as data/test_period_1min.parquet.

SF-LIFE Acknowledgment

SF-ANOMALIES is built on SF-LIFE, which is described in:

Chanuka Algama, Taylor Anderson, Henrique Ferraz de Arruda, Andrew Crooks, Nathan Holt, Erfan Hosseini Sereshgi, John Hunter, Hamdi Kavak, Lance Kennedy, Yueyang Liu, Dieter Pfoser, Sandro Martinelli Reia, Doug Taylor, Mauryan Uppalapati, Boyu Wang, Carola Wenk, and Andreas Züfle. SF-LIFE: A Large-Scale Simulated Movement Dataset for the San Francisco Bay Area. arXiv:2606.00430, 2026.

Downloads last month
389

Paper for sf-life/SF-Anomalies