Abstract:
Fuel consumption of a vehicle depends on several internal factors such as distance,
load, vehicle characteristics, and driver behavior, as well as external factors
such as road conditions, traffic, and weather. Moreover, not all of these factors are
easily obtainable for the fuel consumption analysis. Therefore, fuel-fraud is relatively
easier to conceal; thus, considered a significant threat to the fleet industry
by managers. This research model and evaluate the fuel consumption of fleet vehicles
based on vehicular data and suggest suitable process improvement actions
to improve the fuel economy. We first model and predict the fuel consumption to
identify possible frauds. We considered a case where only a subset of the factors
mentioned above is available as a multivariate time series from a long-distance
public bus. An evaluation of several machine learning techniques revealed that
Random Forest could predict fuel consumption with 95.9% accuracy. To verify
the detected cases of possible fuel fraud, we propose to use different indicators
such as speed profile, the frequency of harsh events, total idle time, and
day of the week. Further, we propose a solution to promote fuel-efficient driving
through real-time monitoring and driver feedback. A classification model, derived
from historical data, identifies fuel inefficient driving behaviors in real-time. The
model considers both the driver-dependent and environmental parameters such
as traffic, road topography, and weather in determining driving efficiency. If an
inefficient driving event is detected, a fuzzy logic inference system is used to determine
what the driver should do to maintain fuel-efficient driving behavior. The
decided action is conveyed to the driver via a smartphone in a nonintrusive manner.
We demonstrate that the proposed classification model yields an accuracy of
85.2% while increasing the fuel efficiency up to 16.4%.
