================================================================ ML-Ready Stop-Level Bus Travel Time Dataset P211 and P411 Bus Services, Johor Bahru, Malaysia 2025 Training Set and 2026 Cross-Year Test Set ================================================================ Version: 1.0 Date: 2025 Licence: Creative Commons Attribution 4.0 (CC-BY 4.0) Repository: University of Glasgow Enlighten Research Data DOI: 10.5525/gla.researchdata.2454 ================================================================ AUTHORS ================================================================ Sye Loong Keoh School of Computing Science, University of Glasgow, UK Corresponding author: SyeLoong.Keoh@glasgow.ac.uk ORCID: 0000-0003-3640-5010 ================================================================ ASSOCIATED PUBLICATION ================================================================ Keoh, S.L. et al. (2026). Machine Learning-Based Bus ETA Prediction: Cross-Year Generalisation and Training-Window Selection in an IoT-Enabled Deployment. [Under review] ================================================================ FUNDING AND ACKNOWLEDGEMENTS ================================================================ This dataset was collected as part of the ASEAN ICT Virtual Organisation (IVO) project: "An IoT-based Public Transport Data Collection and Analytic Framework using Bluetooth Proximity Beacons" Funding: UTM Grant No: R.J130000.7606.4B856 UTM Matching Grant: Q.J130000.3006.04M36 Supported by NICT (Japan) Data collection conducted in partnership with: Johor Public Transport Corporation (Perbadanan Pengangkutan Awam Johor, PAJ) Johor Bahru City Council (MBJB) Iskandar Puteri City Council (MBIP) Kulai Municipal Council (MPKu) ================================================================ DATASET OVERVIEW ================================================================ This dataset contains machine learning-ready stop-level bus travel time records derived from two data sources: (1) Bluetooth Low Energy (BLE) proximity sensing using Estimote iBeacon units mounted on bus dashboards, detected by Raspberry Pi units installed at bus stops. (2) Global Positioning System (GPS) fleet tracking provided by PAJ via a live GPS feed from the bus fleet. Two bus services are covered: P211: Taman Universiti <-> Larkin Sentral Distance: 22.3 km | Stops: 26 per direction Frequency: ~14 return trips daily Typical journey time: 35-45 minutes P411: Kulai <-> Larkin Sentral Distance: 31 km | Stops: ~31 per direction Frequency: ~5 return trips daily Typical journey time: 45-60 minutes Both services share a 14 km corridor between Skudai and Larkin Sentral along Johor Bahru's main trunk road. Data collection periods: 2025 (Training/validation): January - December 2025 2026 (Cross-year test): January - June 2026 Data source by period: P211: BLE detection (Jan-May 2025), GPS (Jun-Dec 2025) P411: BLE detection (Jan-Nov 2025), GPS (Dec 2025) All 2026 data: GPS fleet feed (PAJ) ================================================================ FILES IN THIS DATASET ================================================================ A. ML-READY TRAINING DATASETS (2025 full year) ----------------------------------------------- p211_ml_ready_route8_outbound.csv P211 Route 8 — Outbound (Taman Universiti → Larkin Sentral) Records: 68,744 | Date range: 2025-01-01 to 2025-12-31 p211_ml_ready_route9_inbound.csv P211 Route 9 — Inbound (Larkin Sentral → Taman Universiti) Records: 70,900 | Date range: 2025-01-01 to 2025-12-31 p411_ml_ready_route6_outbound.csv P411 Route 6 — Outbound (Kulai → Larkin Sentral) Records: 33,254 | Date range: 2025-01-01 to 2025-12-31 p411_ml_ready_route7_inbound.csv P411 Route 7 — Inbound (Larkin Sentral → Kulai) Records: 39,970 | Date range: 2025-01-01 to 2025-12-31 B. ML-READY CROSS-YEAR TEST DATASETS (2026 H1) ----------------------------------------------- p211_2026_ml_ready_route8_outbound.csv P211 Route 8 — Outbound, 2026 test set Records: 27,292 | Date range: 2026-01-01 to 2026-06-30 p211_2026_ml_ready_route9_inbound.csv P211 Route 9 — Inbound, 2026 test set Records: 28,878 | Date range: 2026-01-01 to 2026-06-30 p411_2026_ml_ready_route6_outbound.csv P411 Route 6 — Outbound, 2026 test set Records: 16,186 | Date range: 2026-01-01 to 2026-06-30 p411_2026_ml_ready_route7_inbound.csv P411 Route 7 — Inbound, 2026 test set Records: 21,658 | Date range: 2026-01-01 to 2026-06-30 Total ML-ready records across all 8 files: 306,882 C. SPATIAL REFERENCE FILES --------------------------- bus_stops_reference.csv Geographic coordinates of all 75 unique bus stops. route_stop_sequence.csv Ordered stop sequence for each of the 4 route directions. D. ROUTE GEOMETRY FILES ----------------------------------------------------- route_geometry_p211_r8_outbound.geojson — P211 Route 8 road geometry route_geometry_p211_r9_inbound.geojson — P211 Route 9 road geometry route_geometry_p411_r6_outbound.geojson — P411 Route 6 road geometry route_geometry_p411_r7_inbound.geojson — P411 Route 7 road geometry These GeoJSON files contain the actual road polyline geometry for each route. They use WGS84 coordinate system (EPSG:4326). ================================================================ COLUMN DESCRIPTIONS (ML-ready CSV files) ================================================================ route_id Bus route identifier: 6, 7, 8, or 9 date Date of observation (YYYY-MM-DD format) time Confirmed arrival time at destination stop (HH:MM:SS) latitude GPS latitude of stop position at time of arrival (decimal degrees, WGS84) longitude GPS longitude of stop position at time of arrival (decimal degrees, WGS84) time_bin 10-minute time slot index derived from arrival time Range: 0-143, where 0 = 00:00-00:10, 143 = 23:50-00:00 Formula: time_bin = (hour * 60 + minute) // 10 distance_from_stop Road geometry distance between from_stop_id and cur_stop_id along the route polyline (metres) Computed using perpendicular foot projection onto route shape cur_stop_id Destination stop identifier (integer, 1000-series) Links to bus_stops_reference.csv and route_stop_sequence.csv route_order Sequential position of cur_stop_id within the route (1 = first stop, increases toward terminal) time_taken *** TARGET VARIABLE *** Travel time from from_stop_id to cur_stop_id (seconds) Includes dwell time at the origin stop (from_stop_id) For interpolated segments: estimated proportionally by road geometry distance from_stop_id Origin stop identifier (integer, 1000-series) prev_stop_time Confirmed departure time from origin stop (HH:MM:SS) For BLE data: actual detection timestamp For GPS data: closest GPS fix to stop coordinates ================================================================ COLUMN DESCRIPTIONS (bus_stops_reference.csv) ================================================================ stop_id : Unique stop identifier (1000-series integer) latitude : Median GPS latitude across all observations (decimal degrees, WGS84) longitude : Median GPS longitude across all observations (decimal degrees, WGS84) Total stops: 80 Geographic coverage: Johor Bahru, Malaysia Latitude range: 1.4868 N to 1.6626 N Longitude range: 103.5986 E to 103.7427 E ================================================================ COLUMN DESCRIPTIONS (route_stop_sequence.csv) ================================================================ service : Bus service (P211 or P411) route_id : Route direction (6, 7, 8, or 9) route_order : Sequential stop position within route stop_id : Stop identifier (links to bus_stops_reference) P211 Route 8 (Outbound): 25 stops First: 1083 (Taman Universiti area) Last: 1076 (Larkin Sentral) P211 Route 9 (Inbound): 28 stops First: 1077 (Larkin Sentral area) Last: 1084 (Taman Universiti area) P411 Route 6 (Outbound): 30 stops First: 1000 (Kulai area) Last: 1076 (Larkin Sentral area) P411 Route 7 (Inbound): 35 stops First: 1077 (Larkin Sentral area) Last: 1071 (Kulai area) Shared corridor (stops on both P211 and P411): 33 stops ================================================================ DATA PROCESSING PIPELINE ================================================================ The raw BLE and GPS data underwent the following processing steps before producing the ML-ready datasets: Step 1 — Data cleaning - Removed GPS jumping (coordinates outside route geometry) - Removed duplicate records (dwell time at stops) - Corrected route direction misclassifications using cumulative projected distance along route polyline - Removed records where time_taken < 0 or > 7,200 seconds Step 2 — Stop-to-stop duration computation - Arrival at stop defined as GPS coordinate closest to stop location (Haversine distance, minimum threshold) - Duration = confirmed_arrival_time - confirmed_departure_time - For BLE: timestamps are actual BLE detection events - For GPS: timestamps are interpolated from nearest fix Step 3 — Multi-stop interpolation Where bus confirmed at stop A and later at stop B with unobserved intermediate stops s_(a+1)...s_(b-1): estimated_duration(s_j, s_(j+1)) = (T_(a→b) / D_(a→b)) * d_(j→j+1) where T_(a→b) = observed total span duration, D_(a→b) = total road geometry distance, d_(j→j+1) = individual segment road distance Step 4 — Outlier removal Per stop pair (from_stop_id, cur_stop_id): - Removed top 1% of time_taken (anomaly/breakdown records) - Removed bottom 3% of time_taken (GPS errors/data jumping) Note: Interpolated records are included and not separately flagged. Interpolation rates vary by stop pair — see the associated publication for route-level statistics. ================================================================ RECOMMENDED FEATURE ENGINEERING ================================================================ The following features can be derived from the raw columns for machine learning model training: day_of_week = date.dayofweek (0=Monday, 6=Sunday) month = date.month (1-12) quarter = date.quarter (1-4) prev_stop_time_s = seconds since midnight derived from prev_stop_time column Formula: h*3600 + m*60 + s is_school_holiday = 1 if date falls within Johor school holiday period (Kumpulan B schedule), else 0 is_public_holiday = 1 if date is a Malaysia federal or Johor state public holiday, else 0 is_monsoon = 1 if month in {11, 12, 1} (Northeast Monsoon season), else 0 Recommended base feature set (6 features) for cross-year deployment — best cross-year accuracy in associated study: day_of_week, time_bin, from_stop_id, cur_stop_id, prev_stop_time_s, distance_from_stop Target variable: time_taken (seconds) Train/test split used in associated study: 2025 data: 70% training, 30% held-out test random_state = 10 ================================================================ JOHOR SCHOOL HOLIDAY CALENDAR (Kumpulan B) ================================================================ 2025: 01 Jan – 16 Jan (Year-end break from 2024 session) 31 Mar – 04 Apr (Term 1 holidays + Hari Raya) 29 May – 09 Jun (Mid-year holidays) 13 Sep – 21 Sep (Term 2 holidays) 20 Oct – 22 Oct (Deepavali) 20 Dec – 31 Dec (Year-end holidays) 2026 (H1 only): 01 Jan – 11 Jan (Year-end break from 2025 session) 16 Feb – 20 Feb (Chinese New Year) 18 Mar – 29 Mar (Hari Raya + Term 1 holidays) 23 May – 07 Jun (Mid-year holidays) ================================================================ MALAYSIA PUBLIC HOLIDAYS INCLUDED ================================================================ 2025: 29-30 Jan (Chinese New Year), 11 Feb (Federal Territory Day), 01 Mar (Israk Mikraj), 24 Mar (Johor Sultan Birthday), 31 Mar (Hari Raya Aidilfitri), 01 Apr (Hari Raya), 01 May (Labour Day), 02 Jun (Agong Birthday), 07 Jun (Hari Raya Haji), 27 Jun (Awal Muharram), 01 Aug (Johor Heritage Day), 01 Sep (Hari Kebangsaan Johor), 05 Sep (Nuzul Al-Quran), 16 Sep (Malaysia Day), 20 Oct (Deepavali), 25 Dec (Christmas) 2026 (H1 only): 01-02 Feb (Thaipusam / Federal Territory Day), 17-19 Feb (Chinese New Year), 21-23 Mar (Hari Raya Aidilfitri), 01 May (Labour Day), 25-27 May (Hari Raya Haji), 01 Jun (Agong Birthday), 17 Jun (Awal Muharram) ================================================================ CITATION ================================================================ Dataset citation: Keoh, S.L. (2026). ML-Ready Stop-Level Bus Travel Time Dataset. University of Glasgow Enlighten Research Data. DOI: 10.5525/gla.researchdata.2454 Paper citation: Keoh, S.L. et al. (2026) Machine Learning-Based Bus ETA Prediction: Cross-Year Generalisation and Training-Window Selection in an IoT-Enabled Deployment. Engineering Applications of Artificial Intelligence (Elsevier). [Under review] ================================================================ CONTACT ================================================================ Principal Investigator: Prof. Sye Loong Keoh School of Computing Science University of Glasgow Glasgow G12 8RZ, United Kingdom Email: SyeLoong.Keoh@glasgow.ac.uk ORCID: 0000-0003-3640-5010 For data access queries: University of Glasgow Enlighten Research Data: https://researchdata.gla.ac.uk ================================================================ TERMS OF USE ================================================================ This dataset is made available under the Creative Commons Attribution 4.0 International (CC-BY 4.0) licence. You are free to share and adapt the data for any purpose, provided appropriate credit is given. Full licence: https://creativecommons.org/licenses/by/4.0/ The authors make no warranties regarding accuracy or completeness. Users should refer to the associated publication for full methodological details before using the dataset for research. ================================================================