Steel Plates Fault Detection

Project Overview

This project applies machine learning techniques to detect faults in steel plates using a dataset that contains 27 numeric input features and 7 binary fault-type indicators.

Currently, the implementation includes a custom k-Nearest Neighbors (KNN) classifier. Future versions will incorporate additional models such as Decision Trees, Logistic Regression, and Support Vector Machines (SVM(maybe)) to enable comparative analysis and performance evaluation.

Dataset Description

• Source: Steel Plates Faults Dataset (UCI Machine Learning Repository)
• Records: 1,941 samples
• Features:
  • 27 numerical input variables
  • 7 binary columns representing distinct fault types
• Objective: Perform binary classification to predict the presence of a specific fault type (e.g., 'Pastry')

Project Structure

├── data/
│   ├── Faults.NNA              # Raw dataset with no column names
│   └── Faults27x7_var          # List of column names
│   └── Faults.csv              # The complete dataset with the column names
├── main.py                     # Script containing model implementation and evaluation
├── README.md                   # Project documentation

Requirements

Python version 3.7 or above is required.

Required Python Packages

• numpy
• pandas
• matplotlib
• seaborn
• ucimlrepo (needed to download the dataset if it doesn't exist)

You can install these packages using pip:

Creating a Virtual Environment (Recommended)

  python -m venv .venv

  # Then Activate the environment:

  # On Windows:
  venv\Scripts\activate
  or 
  .venv/Scripts/Activate.ps1

  # On macOS/Linux:
  source .venv/bin/activate

  # Install dependences
  pip install -r requirements.txt

How to Run

1. Ensure the dataset files (Faults.NNA and Faults27x7_var) are placed inside a data/ directory. Otherwise install the ucimlrepo library.

2. Run the main script using:

  python main.py

This will:

• Load and preprocess the dataset
• Train and test a machine learning model (initially KNN)
• Output model accuracy
• Generate a visualization of predictions and a confusion matrix

Planned Enhancements

• Support for multiple classifiers (Decision Tree, Logistic Regression, SVM(maybe))
• Hyperparameter tuning and model selection
• Exportable performance reports

Academic Use

This project has been developed for educational purposes in a university setting. All machine learning algorithms are implemented from scratch to provide clarity and a deeper understanding of their internal workings.

Team Information

Team Members:

• Jamal Dassrath (Immatriculation No.: 22301035)
• Love - (Immatriculation No.: 12306406)
• Muhammad Ahtisham Bhatti (Immatriculation No.: 22301502)

Study Course: Artificial Intelligence B.Sc.
Semester: 4th Semester