diff --git a/features/proximity.py b/features/proximity.py
index 1884f29..36c9569 100644
--- a/features/proximity.py
+++ b/features/proximity.py
@@ -5,7 +5,12 @@ import pandas as pd
 from config.models import Participant, Proximity
 from setup import db_engine, session
 
-FEATURES_PROXIMITY = ["freq_prox_near", "prop_prox_near"]
+FILL_NA_PROXIMITY = {
+    "freq_prox_near": 0,
+    "prop_prox_near": 1/2  # Of the form of a / (a + b).
+}
+
+FEATURES_PROXIMITY = list(FILL_NA_PROXIMITY.keys())
 
 
 def get_proximity_data(usernames: Collection) -> pd.DataFrame:
diff --git a/machine_learning/pipeline.py b/machine_learning/pipeline.py
index 56c1d0c..9b3cb72 100644
--- a/machine_learning/pipeline.py
+++ b/machine_learning/pipeline.py
@@ -72,6 +72,11 @@ class SensorFeatures:
             self.df_features_all = safe_outer_merge_on_index(
                 self.df_features_all, self.df_proximity_counts
             )
+            self.df_features_all.fillna(
+                value= proximity.FILL_NA_PROXIMITY,
+                inplace=True,
+                downcast="infer",
+            )
             print("Calculated proximity features.")
 
         if "communication" in self.data_types: