**4. Methodology**

### *4.1. Unobtrusive Stress Detection System with Smart Bands*

Our stress detection system developed in [32] allows users to be aware of their stress levels during their daily activities without creating any interruption or restriction. The only requirement to use this system is the need to wear a smart band. Participants in this study wore the Empatica E4 smart band on their non-dominant hand. The smart band provides Blood Volume Pressure, ST, EDA, IBI (Interbeat Interval) and 3D Acceleration. The data are stored in the memory of the device. Then, the artifacts of physiological signals were detected and handled. The features were extracted from the sensory signals and fed to the machine learning algorithm for prediction. In order to use this system, pre-trained machine learning models are required. For training the models, feature vectors and collected class labels were used.

### 4.1.1. EDA Preprocessing Artifact Detection and Removal Methods

The body sweats when emotional arousal and stress are experienced and, therefore, skin conductance increases [40]. This makes EDA a promising candidate for stress level detection. Intense physical activity and temperature changes contaminate the SC (Skin Conductance) signal. Therefore, affected segments (artifacts) should be filtered out from the original signal. In order to detect the artifacts in the SC signal, we used an EDA toolkit [41] which is 95% accurate on the detection of the artifacts. While developing this tool, technicians labeled the artifacts manually. They trained a machine learning model by using the labels. In addition to the SC signal, 3D acceleration and ST signals were also used for artifact detection. We removed the parts that this tool detected as artifacts from our signals. We further added batch processing and segmentation to this tool by using custom software built-in Python 2.7.
