Modelo multiclasificador aplicando minería de datos para el diagnóstico médico utilizando datos abiertos

Abstract

In order to understand, analyze and eventually make use of a massive amount of data, a multidisciplinary approach has been proposed: data mining, it consists in the process of identifying patterns from massive databases. Data mining can be used in several fields: mobile computing, web mining, financial analysis, crime analysis, engineering, management and medicine. In medicine, data mining techniques can be used by researchers for the diagnosis and prediction of sundry diseases. Researchers have widely used several statistical analysis and data mining techniques to improve the accuracy in the diagnosis of the different diseases in medical care. The use of individual data mining techniques reveals an acceptable level of accuracy to determinate the diagnosis of the disease. Sets or combinations of classifiers have produced promising results, improving accuracy, reliability and, most importantly, extensive coverage in many different and practical applications. In this study, we evaluated the application of feature selection methods and the use of a set of three classifiers based on three different mixing strategies using open diabetes and breast cancer databases contained in the ICU data repository with the goal of supporting decision-making in medical diagnosis. The effectiveness of the combinations was measured using three commonly accepted metrics, as well as the response times in the construction of the models. Analysis of the results reveal that the sensitivity, specificity and percentage of classification are higher compared to the individual techniques and provides the basis for establishing an improved multi-classifier model.