Learning Models

Dataset

The models use the public-access driving dataset POLIDriving. After feature selection, the following attributes were selected: observation hour, speed, rpm, acceleration, throttle position, engine temperature, engine load value, heart rate, current weather, visibility, precipitation, accidents onsite, design speed, accidents time, and risk level.

Libraries

Those models use Concrete ML, a privacy-preserving machine learning (PPML) set of tools based on fully homomorphic encryption (FHE), to convert the learning model to its FHE equivalent.

Models

Considering the available builtin models on Concrete ML, the following PPML models were created.

• Decision tree
• Random forest
• Gradient boosting
• Neural network

Configurations

Data file format

Data files contain the following attributes.

#	Attribute	Class	Units	Data source
1	time	Timestamp		Vehicle data
2	speed	Numeric	km/h	Vehicle data
3	revolutions per minute	Numeric	rpm	Vehicle data
4	acceleration	Numeric	m/s2	Vehicle data
5	throttle position	Numeric	%	Vehicle data
6	engine temperature	Numeric	C	Vehicle data
7	system voltage	Numeric	volts	Vehicle data
8	distance traveled	Numeric	km	Vehicle data
9	engine load value	Numeric	%	Vehicle data
10	latitude	Numeric		Vehicle data
11	longitude	Numeric		Vehicle data
12	altitude	Numeric	m	Vehicle data
13	id vehicle	Numeric		Vehicle data
14	heart rate	Numeric	bpm	Driver's data
15	body temperature	Numeric	C	Driver's data
16	id driver	Numeric		Driver's data
17	current weather	Categorical		Weather data
18	has precipitation	Boolean		Weather data
19	is day time	Boolean		Weather data
20	temperature	Numeric	C	Weather data
21	wind speed	Numeric	km/h	Weather data
22	wind direction	Numeric		Weather data
23	relative humidity	Numeric	%	Weather data
24	visibility	Numeric	km	Weather data
25	uv index	Numeric		Weather data
26	cloud cover	Numeric		Weather data
27	ceiling	Numeric	m	Weather data
28	pressure	Numeric	mb	Weather data
29	precipitation	Numeric	mm	Weather data
30	accidents on site	Numeric	deaths	Traffic accidents
31	design speed	Numeric	km/h	Road geometrics characteristics
32	accidents time	Numeric	deaths	Road geometrics characteristics

Publication

If you use POLIDriving in your research, please cite it as follows.

@article{marcillo2024polidriving,
title={POLIDriving: A Public-Access Driving Dataset for Road Traffic Safety Analysis},
author={Marcillo, Pablo and Arciniegas-Ayala, Cristian and Valdivieso Caraguay, {'A}ngel Leonardo and Sanchez-Gordon, Sandra and Hern{'a}ndez-{'A}lvarez, Myriam},
journal={Applied Sciences}, volume={14},
number={14},
pages={6300},
year={2024},
publisher={MDPI}
}

Downloads

The size of POLIDriving is about 150 MB.

Contact

For questions or suggestions, please contact [email protected]