Q.1. In HW 5, using the lending_2014_ver1_raw.jmp dataset, you built a logistic regression model to predict charged-off (defaulted) loans. Now using the same data, build four different models predicting default using Neural Nets. (1) Run the first neural net model using 1 hidden layer, 3 nodes with the TanH activation function and the squared penalty function in the fitting options. (2) Run the second neural net model using 1 hidden layer, 6 nodes using the TanH activation function and the squared penalty function in the fitting options. (3) Run the third neural net model using 2 hidden layers, 3 nodes each with the TanH activation function. Use the squared penalty function in the fitting options. (4) Run the fourth neural net model on the data using 2 hidden layers, 9 nodes each with TanH, Linear and Gaussian function. Use the squared penalty function in the fitting options. For cross-validation, use the same validation column (with seed 888) that you used in HW 5. Compare and contrast these four models based on the AUC, complexity and run time. Specifically, compare your first and fourth model based on the AUC, complexity and run time. What are your takeaways? Also, comment on which is your best Neural Net model and how does it compare to your best model from HW 5? As always, please show the relevant tables and charts from your output. As in HW 5, the outcome variable to be modelled is loan status. The good news is that in HW 5 you have already done all the data prep (cleaning, variable creation etc.) You created a subset of the dataset to keep only charged-off and paid off loans. You have also changed data types as required, for e.g. earliest_cr_line and last_payment_d, to create features like length of credit etc. For the attributes to include in your two model, please use the same attributes you used in HW 5 baseline model - attributes that are available at the time of investment, and attributes in the historical credit data that are not updated over time. Remember, just as in HW 5, borrowers' credit attributes at the time the loan was issued and their personal attributes are the most likely candidates for your model. Also, in addition to the lecture videos for Neural Nets that I posted on HuskyCT, please refer to the JMP video. https://www.jmp.com/en_ph/events/ondemand/mastering-jmp/neural-networks.html