You're developing a deep learning model to predict the likelihood of a customer making a purchase on your ecommerce site. Initial evaluations show the model is overfitting with both training and test data. What steps can you take to improve the model's accuracy on new data?