Publication Date

Spring 2026

Degree Type

Master's Project

Degree Name

Master of Science in Computer Science (MSCS)

Department

Computer Science

First Advisor

Sayma Akther

Second Advisor

Amith Kamath Belman

Third Advisor

Rajat Kabra

Keywords

Wearable Sensing, Alcohol Consumption Detection, Motion Sensors, IMU, Transdermal Alcohol Concentration, Machine Learning, Gesture Recognition, Time-Series Analysis, Subject Split Evaluation

Abstract

his project presents a multi-scale alcohol consumption behavior detection framework using wearable motion sensors. It uses two publicly available datasets with different label structures: the Bar Crawl dataset, which contains smartphone accelerometer data and delayed transdermal alcohol concentration readings, and a wrist-worn IMU dataset with sample-level drinking and non-drinking labels. For the Bar Crawl dataset, the project uses a backward-search episode inference pipeline because gesture-level labels are unavailable and TAC responses are delayed. For the IMU dataset, the project applies supervised classification using subject-based splits, SMOTE oversampling, threshold tuning, and temporal smoothing. The results show stronger performance in semi-controlled settings, with the best F1-score of 0.820, while free-living detection remains more challenging, with the lowest best-condition F1-score of 0.475. Overall, the project shows that the correct analysis strategy depends strongly on the label structure and sensing conditions of the dataset.

Download

Available for download on Saturday, May 22, 2027

Share

COinS