A Review of Evaluation Metrics in Machine Learning Algorithms

With the increase in the adoption rate of machine learning algorithms in multiple sectors, the need for accurate measurement and assessment is imperative, especially when classifiers are applied to real world applications. Determining which are the most appropriate evaluation metrics to effectively assess and evaluate the performance of a binary, multi-class and multi-labelled classifier needs to be further understood. Another significant challenge impacting research is that results from models that are similar in nature cannot be adequately compared if the criteria for the measurement and evaluation of these models are not standardized. This review paper aims at highlighting the various evaluation metrics being applied in research and the non-standardization of evaluation metrics to measure the classification results of the model. Although Accuracy, Precision, Recall and F1-Score are the most applied evaluation metrics, there are certain limitations when considering these metrics in isolation. Other metrics such as ROC\AUC and Kappa statistics have proven to provide additional insightful into the effectiveness of an algorithms adequacy and should also be considered when evaluating the effectiveness of binary, multi-class and multi-labelled classifiers. The adoption of a standardized and consistent evaluation methodology should be explored as an area of future work.