HUMAN STRESS DETECTION AND PREDICTION USING ANN
Abstract : The present study aims to elucidate the main variables that increase the level of stress at the beginning of military conscription service using an artificial neural network (ANN)-based prediction model. Random sample data were obtained from one battalion of the Lithuanian Armed Forces, and a survey was conducted to generate data for the training and testing of the ANN models. Using nonlinearity in stress research, numerous ANN structures were constructed and verified to limit the optimal number of neurons, hidden layers, and transfer functions. The highest accuracy was obtained by the multilayer perceptron neural network (MLPNN) with a 6-2-2 partition. A standardized rescaling method was used for covariates. For the activation function, the hyperbolic tangent was used with 20 units in one hidden layer as well as the back-propagation algorithm. The best ANN model was determined as the model that showed the smallest cross-entropy error, the correct classification rate, and the area under the ROC curve. These findings show, with high precision, that cohesion in a team and adaptation to military routines are two critical elements that have the greatest impact on the stress level of conscripts.
• The performance of the ANNs will not only provide how well stress can be classified from the features derived from the primary signals for stress but also provide an insight to irrelevant features that exist in the feature space.
• The classification accuracy of the anticipated scheme is higher when compared to the existing schemes, owing to the efficient classification and efficient preprocessing.
• Due to the less false negative errors, the sensitivity evaluation of the proposed FSVM is comparatively higher than that of the existing method, as well specificity is also encountered to be higher in FSVM due to its high true positive rate compared to the existing methods.
• In existing approach, stress can be predicted by face reading, interview, and other activities, people are analyses to each other.
• Stress is a major problem facing our world today and it is important to develop an understanding of how an average person responds to stress in a typical activity like reading.
• A genetic algorithm (GA) could help solve these problems by selecting subsets of features for optimizing ANN stress classifications.
• This can lead to the issue where the ratio for number of samples to the number of connections and weights in an ANN could be relatively small, which could affect classifications.
• When experienced for longer periods of time and not controlled, stress has been widely recognized as a major growing concern in our age adversely impacting society due to its potential to cause chronic illnesses (e.g. cardiovascular diseases, diabetes and some forms of cancer) and high economic costs in societies (especially in developed countries.
• Ongoing pressure can bring about genuine medical issue including uneasiness, sleep deprivation, muscle torment, hypertension and a debilitated invulnerable system.
• The quick increment of mental issues or stress has become an extraordinary test to human wellbeing and nature of life.
• This paper proposes and tests a variety of ANNs that can be used to classify stress in reading using a novel set of stress response signals.
• It also proposes methods for ANNs to deal with hundreds of features derived from the response signals using a genetic algorithm (GA) based approach.
• This proposed method ensures the accurate classification outcome which leads to efficient human life saving.
• They propose a novel mind-set acknowledgment structure that can distinguish five force levels for eight distinct kinds of mind-sets like clockwork.
• In addition, a ranking transformation was proposed utilizing self-reports in order to investigate the correlation of facial parameters with a participant's perceived amount of stress/anxiety.
• The proposed designed is helpful for predicting the mental stress of user using the conversation and social media data.
• A slight performance drop between binary stress detection and 3-class emotion classifcation is expected, as multi-class classifcation is a more challenging task than binary classifcation.
• Te performance drop of the traditional machine learning algorithms occurs for both binary stress detection and 3-class emotion classifcation, measuring physiological data from both the chest-worn sensors and the wrist-worn sensors.
• However, adding signals from the ACC sensor should improve the performance of an ideal model, as more data points are being analyzed in the tasks performed.
• Deep neural networks possess key advantages in their capabilities to model complex systems and utilize automatically learning features through multiple network layers.
• Te key reason for using a convolutional neural network was the advantage of parameter-sharing that a convolutional neural network ofers.
• In a convolutional neural network, a small number of flters can be used for feature extraction across entire inputs.
|