ClassAphasia: An Ensemble Machine Learning Network to Improve Aphasia Diagnosis and Determine Severity

In light of medical biases and variability of Aphasia classes, determining the severity of the Aphasia disorder warrants the use of machine learning to properly leverage patient data for inferences. Machine learning solutions are also inherently efficient with regards to time, allowing for medical professionals to create timely treatment plans to mitigate further damage. Neural networks, however, are extremely computationally expensive, and require resources on scale in order to achieve an acceptable level of accuracy for the medical field. Computationally efficient alternatives, namely XGBoosting and Decision Trees, are equipped with drawbacks themselves. However, working to combine these algorithms to create an ensemble machine learning network effectively eliminates these drawbacks. Thus, the ensemble network highlighted within this paper, ClassAphasia, works to create comprehensive algorithm to accurately measure the severity of Aphasia within individuals by combining the XGBoosting and decision tree algorithms. The ClassAphasia algorithm utilizes data drawn from the Moss Aphasia Psycholinguistics Project Database (MAPPD), which contains 87,673 biometric voice data samples [5]. 60% of the data was split for training the algorithm, while 20% was used for validation and testing each. ClassAphasia achieved an accuracy of 100% on its testing set. Leveraging this algorithm for use in the medical field could assist in speech therapy of millions.