Data Mining Project - Group B

This repository contains the Data Mining Project undertaken by Jiaying Cheng and Ziyi Liu at the University of Kassel for the Intelligent Embedded Systems course in Data Mining. The project explores data analysis, clustering, and classification using various data mining techniques.

Table of Contents

• Project Overview
• Features
• Requirements
• Installation
• Usage
• Project Structure
• Results
• Contributing
• License
• Authors

Project Overview

This project applies a structured approach to data mining, focusing on three main sections:

1. Data Analysis: Including JSON to CSV conversion, data standardization, principal component analysis (PCA), and handling of outliers.
2. Clustering: Using PCA-based feature selection and metrics like the Silhouette Coefficient and Davies-Bouldin Index to determine optimal cluster numbers.
3. Classification: Implementing and evaluating multiple classification models, such as SVM, Multivariate Bernoulli Models, Gaussian Mixture Models, and neural networks.

Each section employs Python and popular data science libraries to preprocess data, conduct analysis, and evaluate models.

Features

• Data Analysis:
  • Conversion from JSON to CSV
  • Standardization of features
  • Principal Component Analysis (PCA)
  • Outlier detection and replacement
• Clustering:
  • Selection of optimal cluster numbers
  • C-means clustering with k-means++ initialization
• Classification:
  • Models implemented: SVM, Multivariate Bernoulli, Gaussian Mixture, and Neural Networks
  • Performance evaluation using accuracy metrics

Requirements

To run this project, you need the following libraries:

• Python (>=3.6)
• pandas
• numpy
• scikit-learn
• matplotlib
• PyTorch (for neural network models)

Install the dependencies with:

pip install -r requirements.txt

Installation

1. Clone the repository:

   git clone https://github.com/yourusername/Data-Mining-Project.git
   cd Data-Mining-Project

2. Install required packages.

Usage

1. Navigate to the code directory containing scripts and notebooks.
2. Run analysis, clustering, and classification steps as needed.

   jupyter notebook

Project Structure

Data-Mining-Project/
├── code/               # Contains main scripts and notebooks
├── data/               # Dataset(s) used in analysis
├── results/            # Directory for saving output results
├── README.md           # Project documentation
└── requirements.txt    # List of required packages

Results

Classification Model Performance

Each classification model’s accuracy for different datasets is summarized below:

Model	Dataset x0	Dataset x1	Dataset x2
SVM	79.75%	97.23%	72.39%
Bernoulli	97.04%	96.93%	74.23%
Gaussian Mixture	4.74%	9.29%	0.9%
Neural Network	97.31%	97.62%	84.25%

The neural network and SVM showed strong performance across datasets, with lower results for Gaussian Mixture due to data distribution.

Contributing

Contributions are welcome! Please submit a pull request if you have suggestions or improvements.

License

This project is licensed under the MIT License.

Authors

• Jiaying Cheng
• Ziyi Liu