Computer Science Department, MS Thesis Presentation , Samuel Uche " DUI Detection from Gait using a Multichannel 1DCNN-Attention-BiLSTM Framework "
11:00 a.m. to 12:00 p.m.
DUI Detection from Gait using a Multichannel 1DCNN-Attention-BiLSTM Framework
Samuel Uche
PhD Student
WPI – Computer Science Department
Wednesday, November 20, 2024
Time: 11:00 a.m. - 12:00 p.m.
Location: Fuller Labs 141
Advisor: Prof. Emmanuel Agu
Reader: Prof. Rodica Neamtu
ABSTRACT:
Alcohol intoxication increases the drinker’s Blood Alcohol Content (BAC) and impairs Psychomotor function. Driving under the influence (DUI) of alcohol impairs the driver’s cognitive and motor functions, and caused over 30% of motor vehicle traffic fatalities in 2017. Unobtrusive methods of detection of driver intoxication can reduce cases of DUI . This master’s thesis proposes a novel hybrid Multi-channel 1DCNN-Attention-BiLSTM Framework approach for automated, passive detection of alcohol intoxication from gait data collected from smartphone (accelerometer) sensors.
Challenges addressed include class imbalance that arose due to insufficient amount of intoxicated gait samples. Such class imbalance can bias machine learning models towards the sober samples and hinder the alcohol intoxication detection model’s performance. Raw time-series sensor data was pre-processed using carefully thought out, task-relevant pre-processing steps before classification using a hybrid Multi-channel 1D-CNN-Attention-BiLSTM Framework to determine if a smartphone user is above the legal driving limit (BAC > 0.08). In rigorous evaluation, our proposed approach achieved an accuracy of 93%, outperforming the current state of the art by 9.5%.