From 9947f549f869a05e00eb16d27a4ec2e1779e6480 Mon Sep 17 00:00:00 2001
From: Mustafa CAVUS <45355486+mcavs@users.noreply.github.com>
Date: Sun, 20 Feb 2022 01:41:33 +0100
Subject: [PATCH 1/2] New metrics are added for imbalanced data

Mathews Correlation Coefficient, Balanced Accuracy, Brier Score and log-loss metrics, which are used in case of imbalanced data, are added.
---
 R/model_performance.R | 31 ++++++++++++++++++++++++++-----
 1 file changed, 26 insertions(+), 5 deletions(-)

diff --git a/R/model_performance.R b/R/model_performance.R
index 3418c1a0c..687ca0e63 100644
--- a/R/model_performance.R
+++ b/R/model_performance.R
@@ -97,11 +97,15 @@ model_performance <- function(explainer, ..., cutoff = 0.5) {
     fn = sum((observed == 1) * (predicted < cutoff))
 
     measures <- list(
-      recall    = model_performance_recall(tp, fp, tn, fn),
-      precision = model_performance_precision(tp, fp, tn, fn),
-      f1        = model_performance_f1(tp, fp, tn, fn),
-      accuracy  = model_performance_accuracy(tp, fp, tn, fn),
-      auc       = model_performance_auc(predicted, observed)
+      recall            = model_performance_recall(tp, fp, tn, fn),
+      precision         = model_performance_precision(tp, fp, tn, fn),
+      f1                = model_performance_f1(tp, fp, tn, fn),
+      accuracy          = model_performance_accuracy(tp, fp, tn, fn),
+      auc               = model_performance_auc(predicted, observed)
+      mcc               = model_performance_mcc(tp, fp, tn, fn),
+      brier_score       = model_performance_brier(predicted, observed),
+      log_loss          = model_performance_logloss(predicted, observed),
+      balanced_accuracy = model_performance_bacc(tp, fp, tn, fn)
     )
   } else if (type == "multiclass") {
     measures <- list(
@@ -167,6 +171,23 @@ model_performance_accuracy <- function(tp, fp, tn, fn) {
   (tp + tn)/(tp + fp + tn + fn)
 }
 
+model_performance_bacc <- function(tp, fp, tn, fn) {
+  ((tp / (tp + fn)) + (tn / (tn + fp))) / 2
+}
+
+model_performance_logloss <- function(predicted, observed) {
+  -mean(observed * log(predicted) + (1 - observed) * log(1 - predicted)) 
+}
+
+model_performance_brier <- function(predicted, observed) {
+  mean((predicted - observed) ^ 2)
+}
+
+model_performance_mcc <- function(tp, fp, tn, fn) {
+  ((tp *  tn) - (fp * fn)) / sqrt((tp + fp) * (tp + fn) * (tn + fp) * (tn + fn))
+}
+
+
 model_performance_macro_f1 <- function(predicted, observed) {
   predicted_vectorized <- turn_probs_into_vector(predicted)
   confusion_matrixes <- calculate_confusion_matrixes(predicted_vectorized, observed)

From 1fbfbee48c6c28b84d1963c818e2d8173174cbb9 Mon Sep 17 00:00:00 2001
From: Mustafa CAVUS <45355486+mcavs@users.noreply.github.com>
Date: Mon, 21 Feb 2022 14:28:16 +0100
Subject: [PATCH 2/2] Update model_performance.R

---
 R/model_performance.R | 2 +-
 1 file changed, 1 insertion(+), 1 deletion(-)

diff --git a/R/model_performance.R b/R/model_performance.R
index 687ca0e63..068b7c04b 100644
--- a/R/model_performance.R
+++ b/R/model_performance.R
@@ -101,7 +101,7 @@ model_performance <- function(explainer, ..., cutoff = 0.5) {
       precision         = model_performance_precision(tp, fp, tn, fn),
       f1                = model_performance_f1(tp, fp, tn, fn),
       accuracy          = model_performance_accuracy(tp, fp, tn, fn),
-      auc               = model_performance_auc(predicted, observed)
+      auc               = model_performance_auc(predicted, observed),
       mcc               = model_performance_mcc(tp, fp, tn, fn),
       brier_score       = model_performance_brier(predicted, observed),
       log_loss          = model_performance_logloss(predicted, observed),