diff --git a/Part 3 - Classification/Section 21 - Evaluating Classification Models Performance/README.md b/Part 3 - Classification/Section 21 - Evaluating Classification Models Performance/README.md
index 9ec18c7..36feb2f 100644
--- a/Part 3 - Classification/Section 21 - Evaluating Classification Models Performance/README.md
+++ b/Part 3 - Classification/Section 21 - Evaluating Classification Models Performance/README.md
@@ -12,4 +12,13 @@ Accuracy increases, although we did not train. So accuarcy cannot be always trus
#CAP: Cummulative Accuracy Profile.
Greater area under the curve, the better is the model.
-![](cap.png)
\ No newline at end of file
+![](cap.png)
+
+ap = area between the perfect model and random model.
+ar = area between the good model and random model.
+
+Accuracy Ratio =(AR) ap / ar.
+AR closer to one better is the model.
+![](cm1.png)
+
+Other option is to draw a 50% line and see where it touches our model(good model) on the y-axis and then make judgement accordingly.
\ No newline at end of file
diff --git a/Part 3 - Classification/Section 21 - Evaluating Classification Models Performance/cm1.png b/Part 3 - Classification/Section 21 - Evaluating Classification Models Performance/cm1.png
new file mode 100644
index 0000000..52d1e62
Binary files /dev/null and b/Part 3 - Classification/Section 21 - Evaluating Classification Models Performance/cm1.png differ