Predicción de la calidad de vida mental: dos abordajes diferentes
Loading...
Official URL
Full text at PDC
Publication date
2024
Authors
Advisors (or tutors)
Editors
Journal Title
Journal ISSN
Volume Title
Publisher
Citation
Abstract
Los trastornos mentales, neurológicos y por consumo de sustancias como la esquizofrenia, depresión, epilepsia, demencia y alcoholismo representan el 13% de las enfermedades a nivel mundial, superando a las cardiovasculares y al cáncer. La salud mental es esencial para el bienestar general, y comprender sus factores determinantes es vital para mejorar la calidad de vida.
Este Trabajo de Fin de Máster utiliza métodos de minería de datos para identificar patrones de consumo de alimentos y hábitos nutricionales en la población española, con el objetivo de construir un modelo predictivo de la calidad de vida mental basado en estos hábitos. Estos modelos se desarrollaron utilizando dos abordajes diferentes: ajustando una variable objetivo continua, y ajustando la versión dicotómica de esta misma variable.
Los datos con los que se realizó provienen del estudio observacional transversal DRECE VI (2018), que incluyó 1500 sujetos de Málaga, Murcia, Burgos y Valladolid, de entre 20 y 74 años. En este TFM se utilizó una submuestra de 1083 personas. Los datos fueron anonimizados y modificados por el tutor antes de su análisis.
Se implementaron varios tipos de modelos de aprendizaje automático, incluidos modelos de regresión, modelos basados en árboles como Random Forest y Gradient Boosting, modelos de Support Vector Machine (SVM) y redes neuronales. El análisis reveló que los modelos ajustados a la variable continua no lograron resultados satisfactorios. No obstante, al dicotomizar la variable objetivo, se obtuvieron mejoras significativas que facilitaron la clasificación de la calidad de vida mental en buena o mala. El modelo de ensamblaje, que combina regresión logística y redes neuronales, fue el más efectivo para clasificar la calidad de vida mental.
Los hallazgos indican que la probabilidad de una buena calidad de vida mental aumenta con la edad, los días de actividad física y el caminar. Por el contrario, la probabilidad de una mala calidad de vida mental aumenta con la ingesta de suplementos vitamínicos, horas sedentarias y consumo de ciertos alimentos como pan de molde, leche enriquecida, churros y galletas de mantequilla.
Para futuros estudios, se recomienda utilizar variables predictoras de mayor calidad y relevancia. Si bien la dicotomización de la variable de respuesta produjo resultados aceptables y facilitó la toma de decisiones clínicas, las clases resultantes mostraron una separabilidad limitada. Se sugiere incluir más observaciones y explorar formas alternativas de analizar la calidad de vida mental continua para desarrollar modelos más robustos y precisos.
Abstract: Mental, neurological, and substance use disorders such as schizophrenia, depression, epilepsy, dementia, and alcoholism account for 13% of the global disease burden, surpassing cardiovascular diseases and cancer. Mental health is essential for overall well-being, and understanding its determinants is vital for improving quality of life. This Master’s Thesis uses data mining methods to identify dietary patterns and nutritional habits in the Spanish population, aiming to build a predictive model of mental quality of life based on these habits. These models were developed using two different approaches: adjusting a continuous target variable, and adjusting the dichotomous version of the same variable. The data used comes from the cross-sectional observational study DRECE VI (2018), which included 1500 subjects from Málaga, Murcia, Burgos, and Valladolid, aged between 20 and 74 years. In this thesis, a sub-sample of 1083 individuals was used. The data was anonymized and pre-processed by the supervisor before analysis. Several types of machine learning models were implemented, including regression models, tree-based models such as Random Forest and Gradient Boosting, Support Vector Machine (SVM) models, and neural networks. The analysis revealed that models adjusted to the continuous variable did not achieve satisfactory outcomes. However, dichotomizing the target variable yielded better results, facilitating the classification of mental quality of life as good or poor. The most effective model for classifying mental quality of life was an ensemble model combining logistic regression and neural networks. Findings indicate that the likelihood of good mental quality of life increases with age, physical activity days, and walking. Conversely, the likelihood of poor mental quality of life increases with the intake of vitamin supplements, sedentary hours, and consumption of certain foods such as sliced bread, enriched milk, churros, and butter cookies. For future studies, it is recommended to use higher quality and more relevant predictor variables. While dichotomizing the response variable produced acceptable results and facilitated clinical decision-making, the resulting classes showed limited separability. It is suggested to include more observations and explore alternative methods to analyze the continuous mental quality of life variable to develop more robust and accurate models.
Abstract: Mental, neurological, and substance use disorders such as schizophrenia, depression, epilepsy, dementia, and alcoholism account for 13% of the global disease burden, surpassing cardiovascular diseases and cancer. Mental health is essential for overall well-being, and understanding its determinants is vital for improving quality of life. This Master’s Thesis uses data mining methods to identify dietary patterns and nutritional habits in the Spanish population, aiming to build a predictive model of mental quality of life based on these habits. These models were developed using two different approaches: adjusting a continuous target variable, and adjusting the dichotomous version of the same variable. The data used comes from the cross-sectional observational study DRECE VI (2018), which included 1500 subjects from Málaga, Murcia, Burgos, and Valladolid, aged between 20 and 74 years. In this thesis, a sub-sample of 1083 individuals was used. The data was anonymized and pre-processed by the supervisor before analysis. Several types of machine learning models were implemented, including regression models, tree-based models such as Random Forest and Gradient Boosting, Support Vector Machine (SVM) models, and neural networks. The analysis revealed that models adjusted to the continuous variable did not achieve satisfactory outcomes. However, dichotomizing the target variable yielded better results, facilitating the classification of mental quality of life as good or poor. The most effective model for classifying mental quality of life was an ensemble model combining logistic regression and neural networks. Findings indicate that the likelihood of good mental quality of life increases with age, physical activity days, and walking. Conversely, the likelihood of poor mental quality of life increases with the intake of vitamin supplements, sedentary hours, and consumption of certain foods such as sliced bread, enriched milk, churros, and butter cookies. For future studies, it is recommended to use higher quality and more relevant predictor variables. While dichotomizing the response variable produced acceptable results and facilitated clinical decision-making, the resulting classes showed limited separability. It is suggested to include more observations and explore alternative methods to analyze the continuous mental quality of life variable to develop more robust and accurate models.