International Research Journal of Engineering and Technology (IRJET)
e-ISSN: 2395-0056
Volume: 08 Issue: 04 | Apr 2021
p-ISSN: 2395-0072
www.irjet.net
Survey Paper on Personalized Multitask Learning for Predicting Stress and Mood Neethu V T1 1Student,
Master of Technology, Computer Science & Engg, RIT Engineering College, Kottayam, Kerala, India ---------------------------------------------------------------------***---------------------------------------------------------------------increases susceptibility to infection and illness. SelfAbstract - The idea of estimating mood, stress, and reported health is so strongly related to actual health mental health indicators using unobtrusive data and all-cause mortality, that in a 29-yearstudy it was collected from smartphones and wearables has been found to be the single most predictive measure of garnering increasing interest. Detecting workplace mortality, above even more objective health stress is another growing body of research. A detailed measures such as blood pressure readings. Clearly, study reported that an omnibus model trained to the ability to model and predict subjective mood and detect all people’s mood based on smartphone wellbeing could be immensely beneficial, especially if communication and usage resulted in a prediction such predictions could be made using data collected accuracy of 66 percent. However, if two months of in an unobtrusive and privacy-sensitive way, perhaps labeled data were collected for each person, then using wearable sensors and smartphones. Such a individual, independent personalized models could be model could open up a range of beneficial trained to achieve 93 percent accuracy in mood applications which passively monitor users’ data and classification. Since obtaining two months of training make predictions about their mental and physical data per person can be considered somewhat wellbeing. The predictions could be useful to any unrealistic, the researchers investigated methods for person who might want a forecast of their future training a hybrid model that weights personalized mood, stress, or health in order to make adjustments examples more heavily, which can be used when there to their routine to attempt to improve it. For are fewer labeled training examples per person. While example, if the model predicts that I will be accurately predicting mood and wellbeing could have extremely stressed tomorrow, I might want to a number of important clinical benefifits, traditional choose a different day to agree to review that extra machine learning (ML) methods frequently yield low paper[1]. Unfortunately, modeling wellbeing and performance in this domain. This is because a one-sizemood is an incredibly difficult task, and a highly fits-all machine learning model is inherently ill-suited accurate, robust system has yet to be developed. to predicting outcomes like mood and stress, which Most of the models suffer from a common problem: vary greatly due to individual differences. Therefore, the inability to account for individual differences. by employ Multitask Learning (MTL) techniques to What puts one person in a good mood does not apply train personalized ML models which are customized to to everyone else. For instance, the stress reaction the needs of each individual, but still leverage data experienced by an introvert during a loud, crowded from across the population. MTL account for party might be very different for an extrovert. individual differences provides large performance Individual differences in personality can strongly improvements over traditional machine learning affect mood and vulnerability to mental health issues methods and provides personalized, actionable such as depression. There are even individual insights. To increase the accuracy, can make use of differences in how people’s moods are affected by several features as important for mental health and the weather. Thus, a generic, omnibus machine wellbeing for improving MTL. learning model trained to predict mood is inherently limited in the performance it can obtain. Key Words: Machine learning, Multitask Learning, Mood prediction. Accounting for interindividual variability via MTL can dramatically improve the prediction of these 1. INTRODUCTION wellbeing states: mood, stress, and health. MTL is a type of transfer learning, in which models are Perceived wellbeing, as measured by self-reported learned simultaneously for several related tasks, but health, stress, and happiness, has a number of share information through similarity constraints. important clinical health consequences. Stress © 2021, IRJET
|
Impact Factor value: 7.529
|
ISO 9001:2008 Certified Journal
|
Page 1078