Publication: Análisis de datos útiles en predicción de trastornos emocionales
Loading...
Official URL
Full text at PDC
Publication Date
2019
Advisors (or tutors)
Editors
Journal Title
Journal ISSN
Volume Title
Publisher
Citations
Exportar
Abstract
El trastorno bipolar es una afección mental en la que una persona tiene unos cambios de ánimo muy marcados. Cuando se produce uno de estos cambios se denomina episodio. La finalidad de este trabajo es predecirlos utilizando Machine Learning (Aprendizaje Automático) sobre unos datos que se obtendrán de forma no intrusiva.
La primera fase del trabajo consiste en la obtención de los datos mediante dispositivos tecnológicos que utilizamos en el día a día, como son los smartphones y las pulseras de actividad, y con ellos tratar de predecir episodios en un paciente con trastorno bipolar y detectar si entrará, en un futuro cercano, en un estado de depresión o manía. Con esto el médico sabrá de antemano cuando un paciente puede sufrir un episodio y reaccionar con antelación de la forma que considere oportuna.
En concreto se ha trabajado con señales de voz, obtenidas de repositorios públicos, y con datos de actividad física para identificar estos estados. Las señales de voz servirán para predecir emociones. Esto ayuda a detectar cambios de ánimo, usado como indicador de un posible episodio. Los registros de actividad física también proporcionan información del estado anímico de un paciente. Por ello se consideran que son fuentes relevantes de datos y han sido elegidas para realizar proyecto.
Para la predicción se tratarán los datos a los que después se aplicarán distintos algoritmos de Machine Learning con el objetivo de determinar cuál de estos algoritmos produce mejores y más fiables resultados a la hora de predecir un episodio.
Además de la predicción de crisis, otros resultados que se pueden extraer es la importancia de las distintas variables a la hora de prever si el paciente tendrá o no un episodio próximamente. Con esto se podría profundizar más en la investigación de estos trastornos aislando los factores de influencia más relevantes para la detección de los cambios.
En la presente memoria se describe un análisis exhaustivo del proceso de obtención y preparación de los datos, además de la selección de algoritmos de reconocimiento y detección que tienen mejor precisión. Se discuten los resultados para proporcionar una herramienta de análisis a los especialistas que les ayude en la toma de decisiones.
Bipolar disorder is a mental condition in which a person has very marked mood changes. When one of these changes occurs, it is called an episode. The purpose of this work is to predict them using Machine Learning on data that will be obtained in a non-intrusive way. The first work phase is to obtain data through technological devices that we use daily, such as smartphones and activity wristbands, and with them, try to predict episodes in a patient with bipolar disorder and detect if he will suffer, in the near future, a state of depression or mania. With this, the doctor will know beforehand when a patient may suffer an episode and react in advance in the way he considers appropriate. In particular it has been worked with voice signals, obtained from public repositories, and physical activity data to identify this states. The voice signals will serve to predict emotions. This helps to detect changes in mood, used as an indicator of a possible episode. The physical activity records also provide information of patient mood. Therefore it is considered that they are relevant data sources and have been chosen to carry out the proyect. For the prediction, the data that will be applied to different Machine Learning algorithms will be treated in order to determine which of these produces better and more reliable results when predicting an episode. In addition from the crisis prediction, other results that can be extracted is the importance of the different variables when it comes to predicting whether or not the patient will have an episode soon. This could further deepen the investigation of these disorders by isolating the most relevant factors of influence for the detection of changes. In the present memory an exhaustive analysis of the process of obtaining and preparing the data is described, in addition to the selection of recognition and detection algorithms that have better accuracy. The results are discussed to provide a analysis tool to the specialist that helps them in the decisionmaking.
Bipolar disorder is a mental condition in which a person has very marked mood changes. When one of these changes occurs, it is called an episode. The purpose of this work is to predict them using Machine Learning on data that will be obtained in a non-intrusive way. The first work phase is to obtain data through technological devices that we use daily, such as smartphones and activity wristbands, and with them, try to predict episodes in a patient with bipolar disorder and detect if he will suffer, in the near future, a state of depression or mania. With this, the doctor will know beforehand when a patient may suffer an episode and react in advance in the way he considers appropriate. In particular it has been worked with voice signals, obtained from public repositories, and physical activity data to identify this states. The voice signals will serve to predict emotions. This helps to detect changes in mood, used as an indicator of a possible episode. The physical activity records also provide information of patient mood. Therefore it is considered that they are relevant data sources and have been chosen to carry out the proyect. For the prediction, the data that will be applied to different Machine Learning algorithms will be treated in order to determine which of these produces better and more reliable results when predicting an episode. In addition from the crisis prediction, other results that can be extracted is the importance of the different variables when it comes to predicting whether or not the patient will have an episode soon. This could further deepen the investigation of these disorders by isolating the most relevant factors of influence for the detection of changes. In the present memory an exhaustive analysis of the process of obtaining and preparing the data is described, in addition to the selection of recognition and detection algorithms that have better accuracy. The results are discussed to provide a analysis tool to the specialist that helps them in the decisionmaking.
Description
Trabajo de Fin de Grado, Universidad Complutense, Facultad de Informática, Departamento de Arquitectura de Computadores y Automática, Curso 2018/2019