Analyzing the performance of machine learning techniques: Applications in dance education and the radiation sector

The main problem with the subjectivity of performance evaluation methodology is that there is no standardized scale against which various methodologies could be compared. Hence, it becomes hard to reveal whether the evaluation itself of a certain machine learning algorithm is accurate and fair. The traditional methods of evaluation which use human judgment, like evaluating the performance of dance teachers, are inherently subjective. Different evaluators are likely to produce different judgments and different results based on their opinion of the issue. In the same manner, although machine learning algorithms have already been utilized for the purposes of medical imaging, treatment planning, and radiation detection, it becomes extremely hard to compare them since there are no validated approaches and datasets available for the purpose of evaluation. This research gap will be filled by providing an objective comparison of machine learning techniques applied to dance education and the radiation industry through a comparison analysis tool. A total of 2000 samples of artificially created dance performances were analyzed using 13 biomechanical and temporal predictor variables to compare the accuracy of three classification models, including support vector machines, logistic regression, and random forests. Accuracy, precision, recall, and F1-score were evaluated as criteria to assess the relative performance of each model. Results indicated that the random forest model provided the best overall accuracy at 86.5%, followed closely by logistic regression, then SVM. Overall, the results suggest a need for a systematic approach for assessing performance and demonstrate that machine learning can provide an effective means of measuring performance objectively and accurately in diverse fields.