Methods of handling missing data for deep learning prediction model of earnings in UCI Adult dataset

Given that we plan to use NNets for classification on the Adult dataset, we must handle missing data. This is not necessary for other ML models such as random forest, decision trees, etc.

####Techniques for handling missing data

1. Basic Statistics : Replace the missing data with the mode, mean or median of the feature vector.
2. One-hot : Create an indicator variable to indicate whether or not the feature has missing data.
3. Nearest Neighbor Imputation : Recursively compute the K-Nearest Neighbors of the observation with missing data and assign the mean or median of the K-neighbors to the missing data.
4. Prediction: Use Random Forest to predict the feature with missing data.
5. Factor analysis : Perform eigen decomposition on the design matrix, project the design matrix onto the first two eigen vectors and replace the missing values by the values that might be given by the projected design matrix.

####Code folder missing_data_imputation.py has the code for different imputation techniques Params and result filenames can be generated with gen* files Data can be generated with preprocess_missing_data.py Plots can be generated with plot * modern_net*.py and net*.py train the podels and predict error rate.