The following is data on K's stock price. The company wants to predict the stock price of tomorrow from the stock price of the past few days. Design, learn, and test the optimal artificial neural network model in the following order, and evaluate its predictive power.

Neural network design There are a total of 100 data on the daily stock price. Select the number of input variables and the number of nodes in the hidden layer to find the optimal model. Select the number of nodes (H) of the hidden layer as N/2 H 2N + 1 compared to the number of nodes (N) of the input layer. However, 0.2 and 0.4 are used for the learning rate. First, use the deafault value. (Try 2-3 things)

Learning (90 days of study) The model selected in item 1 is trained using 90 data excluding the last 10 out of a total of 100 data on daily stock prices.

day1, day2, day3, day4, day 5 = day6

Black (10 days black) Test against the stock price of the last 10 days with the learned neural network.

Take the 10-day test result and calculate the average difference (error) (MAE) from the actual value.

MAE= sum(|predicted value-actual value|) * 1/10 (average error of 10)

Exploring the optimal learning cycle Create various models by repeating 1,2,3, and select the model with the least prediction error using the test data after storing the model in 500, 1000, and 1500 iterations for each model.

Submission 1) Comparison of learning outcomes (MAE) of each learned neural network model

2) Structure of optimal neural network and prediction result

3) What is the optimal learning rate and iteration?

4) Result explanation and discussion NN practice-New.doc