Credit risk prediction with and without weights of evidence using quantitative learning models

Figures & data

Figure 1. Variation in Chi-square values across all features in the credit dataset.

Table 1. Chi-square score and p-values for the various features in the credit dataset.

Variable	Chi-square score	p-value
Status	53.97	0.00
Discretised_duration	45.83	0.00
Savings	25.33	0.00
Discretised_amount	23.19	0.00
Credit_history	17.06	0.00
Discretised_age	7.65	0.01
Property	5.73	0.02
Employment_duration	4.02	0.05

Source: Author's calculations.

Figure 2. Cost function minimisation for the logistic regression model.

Figure 3. Log loss for various threshold probabilities for the logistic regression model.

Figure 4. Training and test accuracy of different neighbour values used in the KNN model.

Table 2. Classification scores for all models.

Model	Precision	Recall	F1 score	Jaccard index	Accuracy	AUC	Log loss
Logistic regression with WOE	0.73	0.93	0.82	0.69	0.72	0.60	9.671
Logistic regression	0.85	0.82	0.84	0.72	0.78	0.76	7.599
KNN	0.88	0.69	0.77	0.63	0.73	0.74	–
Decision tree	0.90	0.61	0.73	0.57	0.69	0.73	–
SVM	0.73	0.92	0.81	0.68	0.71	0.59	–

Source: Author's calculations.

Figure 5. Receiver operating characteristic curve for all five models.

Data availability statement

The data used in the paper can be provided on request.