Learn Together With Me (LTWM): Self-Organizing Map (Pt.1)

Case study: Growth of Orange Trees | Topics: Data Mining, Applied Multivariate Statistical Analysis, Clustering Analysis, Statistical Programming| Method: Self-Organizing Map (SOM)

Figure 1. The illustration of Orange Tree (Source: https://www.deviantart.com/dementiarunner/art/orange-tree-trans-png-589969239)

Quote of The Day: “Be the best version of youself and appreciate it. You are precious!”

Mongga: “(knock-knock) Assalamualaikum Warahmatullah Wabarakatuh, Everyone. How do you do? I hope everything goes well. Let’s go to the main topic about. Here is, Self-Organizing Map. Are you interested to learn and want to know more about Self-Organizing Map? In this part, I’ll explain about:

• Introduction: Self-Organizing Map
• Self-Organizing Map Topological Architecture
• Tools for Self-Organizing Map Analysis
• Imporant things in Self-Organizing Map Process

Here we go, readers! Bismillahirahmanirahim :)”

Introduction

Figure 2. Example of Self-Organizing Maps Output (Source: https://en.wikipedia.org/wiki/Self-organizing_map)

Teuvo Kohonen introduced Self-Organizing Map or Self-Organizing Feature Map (SOM = the abbreveation) a.k.a topology-preserving map, Kohonen Map or Kohonen artificial neural networks in 1980s. SOM is part of Artificial Neural Networks, Unsupervised Learning exactly. SOM uses unsupervised learning to make a low-dimensional (usually 2 (two)-dimensional). One of output of SOM goes to (Figure 2.). Yeah, data visualization!! :)

How SOM make a visualization? SOM uses unsupervised learning to discretized stands for the input space of the training samples a.k.a Map. The results of Kohonen Self Organizing Map will indicate the similarity of characteristics between members in the same cluster. In Self-Organizing Map, weight changes are not only performed on the weight of the line connected to the winning neuronode output, but also on the weight of the line to the surrounding neurons. Each output will proceed to a specific input pattern. Then, we can make our final analysis with some visualizations (clustering) and interpret them as our conclusion.

Figure 3. Self-Organizing Map (SOM) Topological Architecture (Source: http://gorayni.blogspot.com/2014/10/self-organizing-maps.html)

Looking on Figure 3., Self-Organizing Map is a network consisting of 2 (two) layers. There are input layer and the output layer. Each neuron in the input layer is connected with each neuron at the output layer. Each neuron in the output layer stands for cluster (the class) of the given input.

Tools

Here are the list of tools we can use to do Self-Organizing Map analysis. Available on Mac, Windows, LINUX/UNIX.

• RStudio Desktop (download and install guidance on URL: https://rstudio.com/products/rstudio/download/)
• RStudio Server (download and install it, follow the guidance on URL: https://rstudio.com/products/rstudio/download/)
• R-project (download and install it, follow the guidance on URL: https://cran.r-project.org/mirrors.html). Choose your country to download and install informations. Indonesia (URL: https://repo.bppt.go.id/cran/)
• Jupyter Notebook (download and install it, follow the guidance on URL: https://jupyter.org/install.html). We can use Anaconda for the extention to run our codes via Jupyter (Julia, Python, and R languanges programming).
• Python (download and install it, follow the guidance on URL: https://www.python.org/downloads/) Additional: TensorFlow in Python.

Important Components in Self-Organizing Map Process

There are three (3) important components in Self-Organizing Map Process (Haykin, 1999):

• Competition

For each input pattern, neurons calculate the value of each discriminant function which gives a basis for the competition. Certain neurons with the smallest value of discriminant function are declared as neuron winner.

• Cooperation

Neuron winners determine the spatial location of the topology environment excited neurons to provide a basis of cooperation in a neuron environment.

• Synaptic Adaption

Excited neurons lower the value of discriminant functions related to the pattern of input through related weight adjustments so that the response of a winning neuron to the next application with the same input pattern will increase.

The second part is about Problem solving use Self-Organizing Map as a method to Analyze about Growth of Orange Trees.

References

• Anis, Y., & Isnanto, R. R. (2014). Penerapan Metode Self-Organizing Map (SOM) untuk Visualisasi Data Geospasial pada Informasi Sebaran Data Pemilih Tetap (DPT).
• Fausett. L.V (1993). Fundamental of Neural Network: Architectures, Algorithm, and Application. Prentice Hall, 1st edition. ISBN-13: 978–0133341867.
• Haykin, S. (1999). Neural Networks: A Comprehensive Foundation. England: Pearson Education. Page. 23, 43–45.
• School of Computer Science, BINUS University. Self-Organizing Map. URL: https://socs.binus.ac.id/2017/03/20/self-organizing-map-som/
• Wikipedia. Self-Organizing Map. URL: https://en.wikipedia.org/wiki/Self-organizing_map

Closing

Thanks for visiting and reading this article. May this article helpful. See you all in second part! :) Wassalamualaikum Warahmatullah Wabarakatuh. Have a nice day and stay safe! Don’t forget: “Physical distancing, wear mask, study in home”.