Table 1.
Exploratory literature search on machine learning to predict withdrawal from prosecution in IPVW cases.
Fig 1.
Left side corresponds to the previous work, whose results are compared with the ones obtained in this work (right side).
Table 2.
Summary of raw and processed dataset.
Table 3.
Grid search process for ANN classifier.
Table 4.
Grid search process for SVM classifier.
Table 5.
Grid search process for RF classifier.
Table 6.
Best model results for each classifier.
Fig 2.
Confusion matrix for the best ANN model.
Fig 3.
Confusion matrix for the best SVM model.
Fig 4.
Confusion matrix for the best RF model.
Table 7.
Variables filtering results obtained for all candidates, classified as positive influence (POS), neutral or minimal influence (NM) and negative influence (NEG).
Table 8.
Evaluation metrics for each classifier with set of ANN’s most important features.
Table 9.
Evaluation metrics for each classifier with set of SVM1’s most important features.
Table 10.
Evaluation metrics for each classifier with set of SVM2 and SVM3’s most important features.
Table 11.
Evaluation metrics for each classifier with set of RF’s most important features.
Fig 5.
Confusion matrix for ANN model trained with the best 7 parameters from the initial the ANN model.
Fig 6.
Confusion matrix for ANN model trained with the best 27 parameters from the initial the SVM1 model.
Table 12.
Evaluation metrics for each classifier with set of previous work’s most important features.
Fig 7.
Confusion matrix for SVM3 model trained with the parameters used in the previous work.
Table 13.
Variable comparison between previous study and the classifiers optimized for this work.
Information presented in fraction mode, where the denominator represents the number of variables used for the row’s system, and the numerator represents the number of variables of the row’s system that matches the column’s system.
Table 14.
Combined classifiers’ results.
Each row shows the results of the classifier trained with the previous work’s variables and the ones that contain the subset indicated in that row.
Table 15.
Evaluation metrics for each classifier with set of previous work’s most important parameters combined with “plans to abandon”.
Table 16.
Evaluation metrics for each classifier with set of previous work’s most important parameters combined with “current questionnaire”.
Fig 8.
Confusion matrix for the final model obtained by the combination of the previous work’s variables and the new variable “plans to abandon”.