| Summary: | Clinical decision-making for patients with multiple acute or chronic diseases (i.e. multimorbidity) is complex. There is often no ’right’ or optimal treatment due to the potentially harmful effects of multiple interactions between drugs and diseases. This makes it necessary to establish trade-offs between the benefits and risks of different treatment strategies. This means also that there may be high levels of risk and uncertainty when making decisions. One factor that can influence how decisions are made under conditions of risk and uncertainty is the decision maker’s personality. The studies of this dissertation used biosignals and eye-tracking methods and developed pointer tracking techniques to monitor human computer interaction to assess, using machine learning techniques, the individual personality of decision makers. Data acquisition systems were designed and prepared to collect and synchronize: 1) physiological data - electrocardiogram, blood volume pulse and electrodermal activity; 2) human-computer interaction data - pointer movements, eye tracking and pupil diameter; 3) decision-making task data; and 4) personality questionnaire’ results. A set of processing tools was developed to ensure the correct extraction of psychophysiologyrelated features that could manifest personality. These features were combined by several machine learning algorithms to predict the Big-Five personality traits: Openness, Conscientiousness, Extraversion, Agreeableness and Conscientiousness. The five personality traits were well modelled by, at least, one of the sets of features extracted. With a sample of 88 students, features from the pointer movements in online surveys predicted four personality traits with a mean squared error (MSE)<0.46. The blood volume pulse responses in a decision-making task trained in a distinct sample of 79 students predicted four personality traits with a MSE<0.49. The application of the personality models based on the pointer movements in the personality questionnaire in a sample of 12 medical doctors achieved a MSE<0.40 for three personality traits. These were the best results achieved in each context of this thesis. The outcomes of this work demonstrate the huge potential of broader models that predict personality through human behaviour, with possible application in a wide variety of fields, such as human resources, medical research studies or machine learning approaches.
|
|---|