The transportation of significant quantities of hazardous materials by rail is a common practice in Canada and globally. While rail transport maintains a generally strong safety record, the occurrence of train incidents remains a possibility. When such incidents involve trains carrying hazardous materials, the potential for release poses a considerable threat to human health and the environment. Prompt and effective response is crucial to minimizing the consequences of such events. Therefore, the capacity for rapid assessment of the likelihood of hazardous material release following a rail incident represents a critical initial step in emergency response protocols. Furthermore, proactive evaluation of potential release scenarios prior to operations can empower operators to implement effective risk mitigation strategies. The Rail HAZMAT Release Predictor is designed to address these critical needs.

The Rail HAZMAT Release Predictor is a web-based application leveraging artificial intelligence methodologies to forecast the probability of hazardous material release in the event of train-related incidents. The machine learning models underpinning this application have been trained using Canadian incident data. Consequently, its predictive capabilities are best suited for application to incidents occurring within Canada.

The primary objective of this application is to equip rail companies, transportation agencies, and emergency management personnel with a tool for the rapid assessment of potential hazardous material releases in the aftermath of a rail incident. This is achieved by providing readily accessible information concerning the involved trains and a predictive analysis of the likelihood of release.

A detailed dataset covering over two decades (January 1999 to May 2023) was created to build predictive models for potential hazardous material releases in rail transport. This dataset incorporates 5,791 rail incidents involving hazmat carriers from the Transportation Safety Board of Canada, enriched with hourly weather data from Open-Meteo, topographical elevation, and regional hazmat transportation volumes from Statistics Canada. The target variable indicates whether a release occurred (1) or not (0).

Time and location-related variables

Variable Name	Definition and Description	Possible Values and Statistics
Day or Night	Occurrence time (Day or night)	Day (58.88%), Night (41.12%)
Province	The geographic region where the accident or incident occurred	Alberta (25.51%), British Columbia (15.53%), Ontario (29.27%), Quebec (11.56%), Others (18.03%)
Elevation	The height (in meters) above sea level at the location of the accident	Mean: 423.37, Std Dev: 395.30, Min: 0, Max: 2328

Activity, track, and train-related variables

Variable Name	Definition and Description	Possible Values and Statistics
Activity Type	The Accident/Incident Case	Inspection and maintenance (25.49%), Proceeding (14.06%), Switching (50.00%), Others (10.45%)
Accident/Incident Type	The type of train activity at the time of the occurrence	Derailment (47.88%), Collision (15.25%), Dg Leaker (32.28%), Others (4.59%)
Total Hazmat Cars Involved	The total number of cars transporting dangerous goods was involved in the occurrence	1 (74.51%), 2 (12.87%), More than 2 (12.63%)
Number of Trains Involved	Number of trains involved	1 (79.68%), More than 1 (20.32%)
Track Type	Type of Track	Yard (72.59%), Main (11.76%), Other Than Main or Yard (15.65%)
Sub-Division Owner	The owner of the subdivision where the occurrence took place	CN – Canadian National Railway Co. (54.65%), CP - Canadian Pacific Railway Co. (29.46%), Others (6%)

Load-related variables

Variable Name	Definition and Description	Possible Values and Statistics
Operation Volume	The total weight of hazmat transported is measured in metric tons for the month	Mean: 2645547.88, Std Dev: 1317424.092, Min: 834365, Max: 6552497
Hazmat-Class	The classification of hazardous materials involved in the incidents	Class 2 (27.18%), Class 3 (33.68%), Class 8 (11.28%), Others (27.86%)

Weather-related variables

Variable Name	Definition and Description	Possible Values and Statistics
Rain	Liquid precipitation of the preceding hour in centimeters, including local showers and rain from large-scale systems	Mean: 0.064, Std Dev: 0.298, Min: 0, Max: 5.7
Snow Fall	Snowfall amount of the preceding hour in centimeters	Mean: 0.015, Std Dev: 0.094, Min: 0, Max: 1.96
Cloud Cover	Total cloud cover as an area fraction (%)	Mean: 47.37, Std Dev: 37.99, Min: 0, Max: 100
Weather Code	Weather conditions are represented as a numeric code (WMO weather interpretation codes)	Clear (32.28%), Nearly clear (20.74%), Partly cloudy (12.50%), Cloudy (18.05%), Drizzle-Not Freezing (0.85%), Rain-Not Freezing (5.13%), Continues Snowfall (2.71%)
Temperature	Air temperature at 2 meters above the ground (°C)	Mean: 5.587, Std Dev: 12.85, Min: -36.4, Max: 33.4
Relative Humidity	Relative humidity at 2 meters above ground (%)	Mean: 70.640, Std Dev: 17.35, Min: 11, Max: 100
Wind Speed	Wind speed at 10 meters above the ground (m/s)	Mean: 12.39, Std Dev: 7.06, Min: 0, Max: 51.8

Target/Output variable

Variable Name	Definition and Description	Possible Values and Statistics
Hazmat Released Indicator	Indicates the release of dangerous goods from train cars	0 (98.26%), 1 (1.74%)

• Logistic Regression
• Neural Network
• Decision Tree