diff --git a/mllib/src/main/scala/org/apache/spark/mllib/tree/loss/LogLoss.scala b/mllib/src/main/scala/org/apache/spark/mllib/tree/loss/LogLoss.scala
index eec2f19939b6f..d44384bbf1cb0 100644
--- a/mllib/src/main/scala/org/apache/spark/mllib/tree/loss/LogLoss.scala
+++ b/mllib/src/main/scala/org/apache/spark/mllib/tree/loss/LogLoss.scala
@@ -60,8 +60,16 @@ object LogLoss extends Loss {
   override def computeError(model: TreeEnsembleModel, data: RDD[LabeledPoint]): Double = {
     data.map { case point =>
       val prediction = model.predict(point.features)
-      // Use log1p since it is more stable than explicitly writing log(1 + exp()).
-      2.0 * math.log1p(math.exp(-2.0 * point.label * prediction))
+      val margin = 2.0 * point.label * prediction
+      // The following are equivalent to 2.0 * log(1 + exp(-margin)) but are more numerically
+      // stable.
+      if (margin >= 0) {
+        2.0 * math.log1p(math.exp(-margin))
+        //math.log1p(math.exp(w))
+      } else {
+        //w + math.log1p(math.exp(-w))
+        2.0 * (-margin + math.log1p(math.exp(margin)))
+      }
     }.mean()
   }
 }