Driving Pattern Analysis, Gear Shift Classification, and Fuel Efficiency in Light-Duty Vehicles: A Machine Learning...

Area of Science:

• Automotive Engineering
• Machine Learning
• Data Science

Background:

• Vehicle gear shifts significantly impact fuel consumption and performance.
• Accurate real-time gear classification is crucial for optimizing driving efficiency.
• Existing methods may lack the precision needed for dynamic driving conditions.

Purpose of the Study:

• To develop an automatic gear shift classification algorithm for M1 category vehicles.
• To analyze vehicle behavior and driving patterns across different gears.
• To accurately predict instantaneous fuel consumption based on gear selection.

Main Methods:

• Utilized onboard diagnostic system (OBD II) and GPS data, including manifold absolute pressure (MAP), revolutions per minute (RPM), and vehicle speed (VSS).
• Applied unsupervised K-means clustering for data segmentation and pattern identification.
• Employed machine learning classifiers such as K-Nearest Neighbors (KNN), decision trees, logistic regression, and Support Vector Machines (SVMs).
• Developed a multiple linear regression model for fuel consumption estimation.

Main Results:

• The Fine KNN model achieved superior performance with 99.7% accuracy, 99.8% precision, and 99.7% recall.
• Higher gears (3, 4, 5) were correlated with reduced fuel consumption compared to neutral gear.
• The fuel consumption model demonstrated a strong fit (R²=0.897) and predicted lower consumption in higher gears.

Conclusions:

• The proposed automatic gear shift classification algorithm, particularly the KNN model, is highly effective for real-time vehicle analysis.
• Optimizing gear shifts based on this algorithm can lead to significant improvements in fuel efficiency.
• Future integration into driver assistance systems (ADAS) and autonomous vehicles holds potential for enhanced performance and efficiency.