Please use this identifier to cite or link to this item:
Title: Optimize the prediction of the driver's state using feature selection techniques on physiological signals
Authors: Nabhan, Elissa
Advisors: Karam, Walid 
Keywords: Autonomous Vehicles, Physiological Signals, Machine Learning, Feature Selection, Optimization
Subjects: Autonomous vehicles
Signal Transduction - Physiology
University of Balamand--Dissertations
Dissertations, Academic
Issue Date: 2022
Technology has always been an essential contributor to several fields, especially when it comes to increasing safety and avoiding accidents.
Autonomous vehicles (AVs) are one of those fields that need technological improvement to assure and increase the level of safety of the driver. Certainly, Artificial Intelligence (AI) is the fundamental function of AVs, and through it, we can guarantee to monitor the driver's state. One of the most efficient and popular ways to detect the driver's status is by measuring physiological signals, such as Electrodermal Activity (EDA), Electrocardiogram (ECG), and Respiration (RESP).
This study aims to investigate the influence of feature selection techniques on the classification models, and to improve their performance to predict the driver's state, tuning their parameters, and comparing results between several experiments. Since feature selection techniques do not have the same effect on all datasets, three experiments got conducted to collect three different datasets and apply the chosen methods to the models. The classification and feature selection parts were implemented simultaneously on every experiment, then improved separately.
The results were satisfactory, and we have concluded that the Sequential Forward selection (SFS) was the most suitable selection technique on physiological data and to improve the model to a score of 1 and 0.9 in two experiments. In addition, we showed the importance of removal of correlated features on the accuracy of the model. Lastly, we have also noticed that selection techniques work better with a higher level of segmentation, meaning having more training samples.
Includes bibliographical references (p. 46-48)
Rights: This object is protected by copyright, and is made available here for research and educational purposes. Permission to reuse, publish, or reproduce the object beyond the personal and educational use exceptions must be obtained from the copyright holder
Ezproxy URL: Link to full text
Type: Thesis
Appears in Collections:UOB Theses and Projects

Show full item record

Record view(s)

checked on Feb 6, 2023

Google ScholarTM


Items in DSpace are protected by copyright, with all rights reserved, unless otherwise indicated.