diff --git a/src/models/neural_networks/ealstm.py b/src/models/neural_networks/ealstm.py
index a091ec2a4..6f8f0b1f2 100644
--- a/src/models/neural_networks/ealstm.py
+++ b/src/models/neural_networks/ealstm.py
@@ -254,22 +254,31 @@ def __init__(
         if dense_features[-1] != 1:
             dense_features.append(1)
 
-        self.dense_layers = nn.ModuleList(
-            [
+        # add linear layer with nonlinear activation functions
+        dense_layers = []
+        for i in range(1, len(dense_features)):
+            dense_layers.append(
                 nn.Linear(
-                    in_features=dense_features[i - 1], out_features=dense_features[i]
+                    # in = size of previous dense layer
+                    in_features=dense_features[i - 1],
+                    # out = size of current dense layer
+                    out_features=dense_features[i],
                 )
-                for i in range(1, len(dense_features))
-            ]
-        )
+            )
+            if i < len(dense_features) - 1:
+                # add a ReLU to all layers except the final layer
+                dense_layers.append(nn.ReLU())
+
+        self.dense_layers = nn.ModuleList(dense_layers)
 
         self.initialize_weights()
 
     def initialize_weights(self):
-
         for dense_layer in self.dense_layers:
-            nn.init.kaiming_uniform_(dense_layer.weight.data)
-            nn.init.constant_(dense_layer.bias.data, 0)
+            # initialise weights for all linear layers
+            if not isinstance(dense_layer, nn.ReLU):
+                nn.init.kaiming_uniform_(dense_layer.weight.data)
+                nn.init.constant_(dense_layer.bias.data, 0)
 
     def forward(
         self,