Meteorological Data Based Detection of Stroke Using Machine Learning Techniques

Using meteorological data, this study compares Machine Learning approaches such as K-Nearest Neighbors, Support Vector Machine, and Artificial Neural Networks for detecting days with a greater probability of stroke incidence in the region of Transylvania, Romania. Being the first to address this problem in Romania, the study contributes to previous research by employing Machine Learning approaches and applying them to meteorological data that also includes air fronts. Furthermore, the study was conducted on data which was collected in a ten-year span (2013-2022). Because the initial dataset had a substantial class imbalance, having the positive class represent 1/20 of the whole dataset, the proposed approaches include dimensionality reduction and clustering techniques. According to the obtained results, the best-performing model is the Support Vector Machine, having an accuracy of 63%, a precision of 70%, and a recall of 63%.