Uso de las técnicas DownSampling y UpSampling para abordar el desequilibrio de datos en la predicción de personas propensas a sufrir accidentes cerebrovasculares

The DownSampling and UpSampling data balancing techniques are used applied to a set related to individuals prone to having a stroke. The purpose of this work is to demonstrate the importance of the application of DownSampling and UpSampling techniques when we find data that present imbalance; making a comparison between the two mentioned techniques and analyzing the behavior of the measures that are calculated in the confusion matrix when the prediction model is created. The data set is composed of 4981 records, of which 4773 belong to the class of those who have not suffered a stroke and 248 to the class that has had one. It was found that for this data set the best technique to treat the imbalance is UpSampling with the largest of its replicas and when the model is going to be evaluated it is important not only to base it on its Accuracy, but also on other measures that result from the confusion matrix, this to achieve a better analysis of the results obtained.