Regression with mini-batch gradient descent was done with a 3-layer neural network to predict median house values ($) in the California 1990 census.
Methods: Regression was done with a 3-layer neural network with He-initialized weights. The model used mini-batch gradient descent with gradient clipping and cross-validation while training. Batch normalization and dropout were also used to facilitate learning and add regularization to the model, respectively. Some missing data analysis was performed pre-modeling.