Classification of Acid-Base Disorders Through Machine Learning

The present research aims to develop a tool in the form of an algorithm that can provide an accurate diagnosis of a patient's acid-base balance without the need for manual calculations by the attending physician. During the research, three different algorithms (Bayesian, KNN and a neural network) were used, tested and compared in order to achieve a reliable result, speeding up the diagnostic process for the patient and reducing the human error that can arise from manual calculations. The results show that the Bayesian algorithm had the lowest performance achieved with an MF1 of 0.7850, followed by the KNN algorithm with an MF1 of 0.8553, the next was the neural network which obtained an MF1 of 0.9711. Finally, an algorithm assembled by the three mentioned above was tested generating an MF1 of 1, which was tested on 70 data samples. This suggests that the design can be used for the classification of acid-base problems.