Reference no: EM132249165

Individual Exercise

Deliverables: Two Files: (1) Submit this lab report with answers to all questions including output screenshots. (2) Submit an R script that contains all commands with comments that briefly describe each commands purpose.

Part 2 - Run an exercise on the vertebral column dataset from column.csv, completing this report and providing the commands, output screenshots, and discussion/interpretation as requested. Ensure that all commands are saved in this report AND in an R script.

For Reference: UCI Machine Learning Repository: Vertebral Column

a. Introduction:

i. Identify the dependent variable and independent variables in the Vertebral Column data set.

ii. Based on what you have learned this week about neural networks, provide a one-paragraph masters-level response describing what you anticipate that the neuralnet algorithm will accomplish for the Vertebral Column data? Be specific about the behavior and structure of neural network model. (80-120 words)

b. Data Pre-Processing: Load the Vertebral Column data into R Studio using the read.csv command (do not use File > Import Dataset > From CSV in the R Studio GUI as this uses read_csv() resulting in significant different variable types).

i. What data pre-processing (if any) does the neuralnet method require for the Vertebral Column data? Include the commands you ran and the output screenshot.

Command(s): >

Output:

c. Neural Network - Running the Method:

i. Run ‘set.seed(12345) and then run the neuralnet() function to build the network storing the results in a variable called ‘nn' with 2 hidden layers and setting linear.output to TRUE. Include the command and discuss the input parameters you used. WARNING: When building your neural network, you may notice that it takes a long time to build or periodically may error out with a failure to converge. If this happens, simply run the command again until it works.

Note: Do not shortcut the independent variable list in your formula with a period, as in do not use ‘class~.' as the formula.

Command: >

Discussion:

ii. Run the command ‘nn$result.matrix'. Include the output screenshot and answer the following questions:

Output:

How many steps were needed to build your neural network?

Describe how the relationships between the independent variables and hidden layers are represented? What about the hidden layers to the dependent variable? (80-120 words)

iii. Run the command ‘nn$net.result[[1]][1:10]'. Include the output screenshot and answer the following question:

Output:

What do each of the ten results above represent? Relate your answer back to the classification of patients from the Vertebral Column dataset. (Hint: patients diagnosed as normal are 0 and those with either disk hernia or spondylolisthesis are 1.)

d. Neural Network - Visualization:

i. Run the plot() command on your neural network ‘nn'. In the space below provide a zoomed screenshot of the plot ONLY so that it is completely visible and all components legible. (Hint: Use the Plots tab Zoom). Include your command, the output screenshot of your plot, and a one-paragraph, masters-level interpretation of all visible aspects of your neural network.

Command: >

Output:

Interpretation: (80-120 words)

e. Neural Network - Evaluate the Model:

i. Run the two-step model evaluation process from the tutorial providing those two commands and the command to display the first 10 values from the variable ‘mypredict' that you create. Include all three commands and the output screenshot.

Commands: >

Output:

ii. Run the table command to build the confusion matrix using ‘mypredict' as the first argument and the Vertebral Column dataset dependent variable as the second. Include the command, output screenshot of your matrix, and answer the following question:

Commands: >

Output:

What is the classification accuracy of your neural network? Provide the complete formula used (i.e. show your work) along with the final percentage (rounded to two decimals places)

f. Neural Network - Running the Method Once More with Different Inputs:

i. Repeat the steps from 2.c, 2.d, and 2.e (Running the Method, Visualization, and Evaluate the Model) but using a different combination of input parameters. At a minimum, you need to change the number of hidden layers and the number of nodes in the hidden layers. Explore the available customizations by reading help(neuralnet) to improve the accuracy of your model. (Hint: Use a vector for hidden = c(x,y) where x and y are the number of hidden layers and nodes.)

All commands from the steps listed must be included in the command block below. You are free to work with and modify your commands in R Studio prior to putting your final set here that ultimately show model improvement.

The only output required is the plot of the neural network.

Commands: >

Output:

ii. What is the classification accuracy for the new neural network you just built? Provide the complete formula used (i.e. show your work) along with the final percentage (rounded to two decimals places).

iii. Compare the classification accuracy from the first neural network run to your final second run. Which has a higher classification accuracy? Provide a one-paragraph, masters-level response that provides reasonable justification for why one is higher than the other. The demonstration of your analysis and depth of understanding are being evaluated above simple right or wrong answers. (80-120 words)

g. Summary:

i. What differences did you observe between the function and behavior of decision tree classification and neural network classification? Support your observations with external research and provide a two-paragraph, masters-level response. (80-120 words)

ii. Which part of this exercise did you find the most challenging and what steps did you take to resolve the challenge?