neural nets

.docx

School

Grand Canyon University *

*We aren’t endorsed by this school

Course

MIS 655

Subject

Information Systems

Date

May 15, 2024

Type

docx

Pages

Uploaded by ProfInternet10544 on coursehero.com

Grand Canyon University – MIS 655 Neural Networks Professor Madson Bryan Howard 2-21-2024

Part 1 Question 1: Neural networks assume the ability to learn and represent complex relationships within data, and the universal approximation theorem suggests that neural networks approximate any continuous function given enough neurons and layers (Nielsen, 2015). Neural networks require enough labeled data to learn meaningful patterns. Insufficient data can lead to overfitting, where the model performs well on training data but fails to generalize to new, unseen data (IBM, n.d). The observations in the dataset are typically assumed to be independent and identically distributed, meaning that one data point does not depend on another Nielsen, 2015). Neural networks are often considered as "black box" models, making it challenging to interpret and understand the reasons behind their predictions (Donges, 2023). Training large neural networks depends on multiple computer resources which become time-consuming, especially when dealing with massive datasets (2023). Furthermore, neural networks are prone to overfitting, however, standardization or regularization techniques are often employed to mitigate this issue 2023). Lastly, neural nets require substantial amounts of labeled data for training, and their performance can degrade when faced with limited or noisy data (2023). Neural networks are suitable for a variety of business problems, especially those involving complex patterns and high-dimensional data. Neural networks are great for image recognition and speech recognition like surveillance, car automation, and more tasks (IBM, n.d.). Furthermore, neural nets excel at natural language processing like chatbots, language translation, automated content generation, and sentiment analysis (IBM). Lastly, neural nets make fraud detection and other transactional recognition easy as they analyze copious amounts of data to identify patterns (IBM). Question 2: Utilizing the R software to analyze datasets, the dividend dataset loaded into R labeled as “divdat” with the dimensions 200 X 6. Below is the screenshot of the summaries of the variables for the dataset.

Question 3: Then the dataset gets partitioned into training and test sets with a 60/40 split which rendered a dataset with 120 rows and eighty rows, respectively. Question 4: Then test set is segregated by the dividend variable to evaluate for accuracy later in the model. Also, the training set and test set were standardized using the scale function R Studio. The test set had the dividend variable removed since it is the target variable of the analysis. Question 5: Here, a neural network is created using the neuralnet function. The formula Div ~ FreeCashflow + EarningsGrowth + DebtEquity + Cap + CR specifies that the variable Div is the target variable, and the others are the input features. The hidden=2 argument indicates that there are two hidden layers in the neural network, and linear.output=FALSE specifies that the output should be transformed using a non-linear activation function. Using the predict function in R, the results of the neural net model using the test variable are stored as a variable labeled nn.results. Then a data frame is created using real and prediction variables, actual was not available in R due to internal conflicts. Real is the test_actual variable and prediction is the nn.results variable. Next, a data frame is created with rounded results from the results frame which rounded the observations to the nearest zero, in this case a 1 or 0.

Your preview ends here

Eager to read complete document? Join bartleby learn and gain access to the full version

Access to all documents
Unlimited textbook solutions
24/7 expert homework help