diff --git a/book/3-classification.tex b/book/3-classification.tex
index 7ebe745..d832339 100644
--- a/book/3-classification.tex
+++ b/book/3-classification.tex
@@ -344,7 +344,22 @@ \subsection{Balanced Accuracy}
 
 % formula
 \begin{center}
-    [FORMULA GOES HERE]
+    \tikz{
+        \node[inner sep=2pt, font=\Large] (a) {
+            {
+                $\displaystyle
+                Balanced-Accuracy = \frac{1}{2} \left( 
+                \frac{\textcolor{nmlred}{TP}}{\textcolor{nmlred}{TP} + \textcolor{nmlgreen}{FN}} +
+                \frac{\textcolor{nmlcyan}{TN}}{\textcolor{nmlcyan}{TN} + \textcolor{nmlpurple}{FP}}
+                \right)
+                $
+            }
+        };
+        \draw[-latex, nmlgreen, semithick] ($(a.south)+(2.1,0.04)$) to[bend left=25] node[pos=1, left] {\color{nmlgreen} false negative} +(-.8, -.5);
+        \draw[-latex, nmlpurple, semithick] ($(a.south)+(5.1,0.04)$) to[bend left=-25] node[pos=1, right] {\color{nmlpurple} false positive} +(.8, -.7);
+        \draw[-latex, nmlred, semithick] ($(a.south)+(1.5, 1.3)$) to[bend right=25] node[pos=1, left] {\color{nmlred} true positive} +(-0.9, .6);        
+        \draw[-latex, nmlcyan, semithick] ($(a.south)+(4.5, 1.3)$) to[bend left=25] node[pos=1, right] {\color{nmlcyan} true negative} +(0.9, .6);
+    }
 \end{center}
 
 It is particularly useful when working with datasets where some classes are significantly underrepresented, ensuring that the performance of all classes is treated equally.