Chemicals Distillation Using Self-organizing Maps

This project utilizes self-organizing maps (SOMs), an artificial neural network renowned for its effectiveness in dimensionality reduction and data visualization. An SOM facilitates the comprehension of intricate data patterns by organizing high-dimensional data into a 2D grid while preserving relationships. Moreover, SOMs are particularly valuable for clustering and exploratory data analysis because they employ a competitive learning process.

The primary objective of this project revolves around the analysis of ceramic samples’ chemical composition through the utilization of SOMs. By employing SOMs, we will effectively group similar ceramics based on their chemical composition to identify meaningful patterns and relationships amongst the diverse ceramic samples.

In this project, we will use the MiniSom library to implement an SOM. The dataset is loaded and preprocessed, including data normalization. We will visualize the trained SOM using heatmaps and provide insights into neuron relationships and ceramic clustering. The final plot will display the ceramics’ names at their respective positions on the SOM, revealing composition-based clusters. Additionally, the influence of neurons will be analyzed and plotted to understand their importance in the clustering process.