|
S/N | TTV subcategory | Reference | TTV | Future work |
|
1 | Feature engineering-induced threats | [L04] | Failure to apply full parameter search | Intend to apply full parameter search in the future |
[L40] | Feature selection does not consider semantics | Evaluation of other feature selection schemes in literature |
[L32] | Choice of feature selection algorithm | Intent to deploy other feature selection algorithms |
[L12] | Threats from feature selection | Checking for effects of imbalance & outliers on feature selection |
|
2 | Human error proneness | [L06], [L11], and [L26] | Varying degrees and kinds of errors | No mention in future works |
[L27] and [L29] |
|
3 | Data/software metrics related threats | [L12], [L21], and [L26] | Restriction of studies to particular software metrics and repositories | Studies intended to try proposed models on other data metric repositories as [L38] intend to address class imbalance in future and [L29] aiming to try commercial metrics repositories |
[L27], [L33], and [L37] |
[L38], [L44], and [L45] |
[L34] |
|
4 | Evaluation-related threats | [L15] | Standard deviation may not be an optimal choice for model evaluation | Other evaluation techniques to be considered except standard deviation |
[L21] | Bias in choice of performance parameters | Root square means method to be adopted in subsequently |
[L38] | AUC & MCC may not be appropriate measures | Trial of other performance evaluation measures |
|