diff --git a/spark/src/main/scala/org/apache/comet/serde/arrays.scala b/spark/src/main/scala/org/apache/comet/serde/arrays.scala
index 8eda097ce6..8241051cf2 100644
--- a/spark/src/main/scala/org/apache/comet/serde/arrays.scala
+++ b/spark/src/main/scala/org/apache/comet/serde/arrays.scala
@@ -22,7 +22,7 @@ package org.apache.comet.serde
 import scala.annotation.tailrec
 import scala.jdk.CollectionConverters._
 
-import org.apache.spark.sql.catalyst.expressions.{And, ArrayAggregate, ArrayAppend, ArrayContains, ArrayExcept, ArrayExists, ArrayFilter, ArrayForAll, ArrayInsert, ArrayIntersect, ArrayJoin, ArrayMax, ArrayMin, ArrayPosition, ArrayRemove, ArrayRepeat, ArraySort, ArraysOverlap, ArraysZip, ArrayTransform, ArrayUnion, Attribute, Cast, CreateArray, ElementAt, EmptyRow, Expression, Flatten, GetArrayItem, IsNotNull, Literal, Reverse, Sequence, Size, Slice, SortArray, ZipWith}
+import org.apache.spark.sql.catalyst.expressions.{And, ArrayAggregate, ArrayAppend, ArrayContains, ArrayExcept, ArrayExists, ArrayFilter, ArrayForAll, ArrayInsert, ArrayIntersect, ArrayJoin, ArrayMax, ArrayMin, ArrayPosition, ArrayRemove, ArrayRepeat, ArraySort, ArraysOverlap, ArraysZip, ArrayTransform, ArrayUnion, Attribute, Cast, CreateArray, ElementAt, EmptyRow, Expression, Flatten, GetArrayItem, IsNotNull, LambdaFunction, Literal, NamedLambdaVariable, Reverse, Sequence, Size, Slice, SortArray, ZipWith}
 import org.apache.spark.sql.catalyst.util.GenericArrayData
 import org.apache.spark.sql.internal.SQLConf
 import org.apache.spark.sql.types._
@@ -730,10 +730,12 @@ object CometArrayFilter extends CometExpressionSerde[ArrayFilter] {
       expr: ArrayFilter,
       inputs: Seq[Attribute],
       binding: Boolean): Option[ExprOuterClass.Expr] = {
-    expr.function.children.headOption match {
-      case Some(_: IsNotNull) =>
-        // Fast path: `array_compact` lowers to `filter(arr, x -> x is not null)`. Use the native
-        // array_compact serde to avoid the per-batch JNI cost of the codegen dispatcher.
+    expr.function match {
+      case LambdaFunction(IsNotNull(v: NamedLambdaVariable), Seq(lambdaVar), _)
+          if v.exprId == lambdaVar.exprId =>
+        // Fast path: Catalyst desugars `array_compact` to `filter(arr, x -> x IS NOT NULL)`, so
+        // restore the native serde here (avoids per-batch JNI). Guard requires the IsNotNull
+        // operand to be the lambda variable itself, not a captured column (#4830).
         CometArrayCompact.convert(expr, inputs, binding)
       case _ =>
         // General lambda: run Spark's own evaluation through the codegen dispatcher so the result
diff --git a/spark/src/test/resources/sql-tests/expressions/array/array_filter.sql b/spark/src/test/resources/sql-tests/expressions/array/array_filter.sql
index c4511e28e4..f7375744f3 100644
--- a/spark/src/test/resources/sql-tests/expressions/array/array_filter.sql
+++ b/spark/src/test/resources/sql-tests/expressions/array/array_filter.sql
@@ -29,3 +29,22 @@ SELECT filter(arr, x -> x >= 0) FROM test_array_filter
 
 query
 SELECT filter(arr, (x, i) -> i > 0) FROM test_array_filter
+
+statement
+CREATE TABLE test_array_filter_captured(arr array<int>, c int) USING parquet
+
+statement
+INSERT INTO test_array_filter_captured VALUES
+  (array(1, NULL, 2), 5),
+  (array(3, NULL, 4), NULL),
+  (array(NULL, NULL), 7),
+  (NULL, 1)
+
+-- Genuine array_compact fast path: IsNotNull on the lambda variable drops the null elements.
+query
+SELECT filter(arr, x -> x IS NOT NULL) FROM test_array_filter_captured
+
+-- Regression for #4830: IsNotNull on captured column `c` is not array_compact and must not drop
+-- the null elements of `arr`.
+query
+SELECT filter(arr, x -> c IS NOT NULL) FROM test_array_filter_captured