Your company has developed a TensorFlow neural network model that includes a large number of neurons and layers. While the model shows excellent performance with the training data, its performance significantly drops when evaluated with new, unseen data. What strategy can you employ to rectify this issue?