Running an Artificial Neural Network (ANN) in RapidMiner Studio #SIDLIT2019

("Artificial Neural Network" by Ahmed Gad on Pixabay, open-source image, no date)

Presentation Description

Computational neural networks were designed and built over decades of work to mimic the human brain in order to process learning through examples. This shows the setting up and running of several artificial neural network (ANN) on educational data, in a software program which enables free educator licensure for non-commercial research projects (in three-year increments).

Artificial Neural Networks (ANNs)

"Multi-layer Neural Network" (by John Salatas, CC BY-SA 3.0, Sept. 10, 2011, on Wikimedia Commons)

An "artificial neural network" (ANN) is a form of machine learning that enables computers to learn by examples by running input data about the examples through neuronal layers that extract different insights about the inputs (text, graphics, and others) in order to enable awareness, predictivity (like categorizations) modeling, and decision-making, among others. Some ANNs are generic ones to handle certain types of data, and others are dedicated ones to solve particular issues.

"Exemplar learning" (or "learning by examples") is harder than top-down learning because the humans and machines have to study a mass of features and determine which are most relevant for particular understandings.

In some ways, this process enables predictivity (and classification) but without full explanatory power. There has to be post-analysis to understand the outcomes and produce the insights. Some degree of mystery may remain in terms of the features being extracted.

ANNs require fairly large datasets to run effectively. The datasets (of examples) also need to be representative of the in-world phenomena. Anything left out of the datasets will skew outcomes. Those aspects of datasets only represented in insufficient numbers may get short-shrift in representation. The ANN discriminatory (ability to differentiate) power differs depending on the data and on the parameters of the ANN runs. (For example, how many layers in the ANN? What types of ANN--for predictivity, for clustering, for applied problem-solving, and others?)

There are some endeavors to help an ANN differentiate with more discernment (and not get confused by noise), beyond tuning the ANN. Some of the input data may be edited to provide variations, so that the ANN may train to understand even across the variations.

ANNs may be set up as data-informed models to optimize approaches to some types of data.

ANNs can learn with both labeled and unlabeled data. They may be tested and validated against labeled data, percentage data, and other empirics.

ANNs may be run over imagesets and videosets as well. Images are decomposed to patterned visual grids, and videos are decomposed to still images decomposed to patterned visual grids. The initial layers of the ANN captured micro-aspects of the still and stilled images...and then capture higher-level integrated image elements in later rounds(layers). The discrimination between images for classifications improves with larger imagesets, modified images (angling images in the set, zooming, changing the visual perspective), and other variations. Human domain knowledge of the challenge is important to properly apply this and other forms of machine learning.

General Sequence

  1. Design research.
  2. Conduct research.
  3. Collect data.
  4. Clean data.
  5. Label data for supervised machine learning to test predictivity of ANN.
  6. Go with unlabeled data for unsupervised machine learning to test clustering using ANN.
  7. Start RapidMiner (RM).
  8. Load data.
  9. View descriptive statistics about the data.
  10. Select the appropriate type of ANN. Set up the sequence in RM with context-sensitive help. Set up the parameters.
  11. Run.
  12. Analyze data.
  13. Analyze visuals.
  14. Analyze text-based reportage.
  15. Re-run based on different parameters.
  16. Copy out the data visualization images and other desirable elements.

A Few Walk-throughs

A Sonar Dataset and Predictivity between Stones and Mines
Run Sonar Dataset to View Data
Open-Source Sonar Dataset
Design for an ANN Model and Validation - Sonar Dataset
An ANN from Sonar Dataset between "Stones" and "Mines"
Improved Neural Net ANN - Sonar Dataset

Several artificial neural network processes will be run over educational data.

The Golf Dataset and Predictivity between Play and Non-Play
Open-Source Golf Dataset
Design for a Deep Learning ANN Model and Cross-Validation - Golf Dataset
Performance Results - Golf Dataset
Performance Vector, Deep Learning ANN - Golf Dataset
The Titanic Dataset and Predictivity between Survival and Nonsurvival with a Deep Learning ANN
Open-Source - Titanic Dataset
Design for a Deep Learning ANN Model and Cross-Validation - Titanic Dataset
Deep Learning ANN Performance - Titanic Dataset (with Particular Parameters)
Deep Learning Artificial Neural Network Details - Titanic Dataset
The Labor Negotiations Dataset and Class Predictivity with an ANN
Open-Source - Labor Negotiations Dataset
Descriptive Statistics - Labor Negotiations Dataset
Interactive Pivot Table - Labor Negotiations Dataset
Process - Labor Relations
Deep Learning Results - Labor Relations Dataset

Gad, A. (n.d.) "Artificial Neural Network." Pixabay. Retrieved May 9, 2019, from https://pixabay.com/illustrations/artificial-neural-network-ann-3501528/.

"Multilayer Neural Network." (2011, Sept. 10). Wikimedia Commons. Retrieved May 9, 2019, from https://upload.wikimedia.org/wikipedia/commons/3/30/Multilayer_Neural_Network.png.


Dr. Shalin Hai-Jew

  • ITS, Kansas State University
  • 785-532-5262
  • shalin@ksu.edu
Created By
Shalin Hai-Jew

Report Abuse

If you feel that this video content violates the Adobe Terms of Use, you may report this content by filling out this quick form.

To report a copyright violation, please follow the DMCA section in the Terms of Use.