Figure 9 is a comparison of the machine learning methods described in this web site. The headings in the table show advantageous modelling attributes.
Resistance against data outliers - A method's robustness at eliminating data outliers in dataset which are highly atypical projects.
Software datasets (metric data) is frequently contaminated with data outliers from a number of causes, including measurement errors and unusual projects. The existence of such data points can lead to inaccurate predictions. Methods that can prevent data outliers from exerting influences in the prediction.
Explains Output - The capability for a
user to see how a model arrives at its predicted output.
Outputs usually fall into three categories:
White
Box - The process is very transparent and well
understood by the user.
Grey
Box - The process is partially visible and
reasonably understood by the user.
Black
Box - The process is hidden from the user.
Suitable for small datasets - Small datasets are very problematic when predicting estimation. This heading assess whether the method is viable when using small datasets.
Reasoning Capability - This assess whether a model has viable reasoning capabilities. For example, in the analogy method you can look back at past projects and question whether a project is really a true analogous project.
Adaptive - This addresses whether the method can accept new data without regenerating the whole estimation process.
