COM6018 Data Science with Python

Week 7: Introducing Scikit-Learn

Jon Barker

Copyright © 2023–2025 Jon Barker, University of Sheffield. All rights reserved.

In this lab

Using Scikit-Learn to explore a classification task

• Splitting a dataset into training and test sets
• Building a k-Nearest Neighbours classifier
• Tuning the hyperparameters of a classifier
• Using Leave-One-Out Cross-Validation to evaluate a classifier
• Using a confusion matrix to look at the patterns of errors.

The Task

• We will be building a land-mine detection system.
• Measurements are taken from a metal detector.
• We predict whether the object is a land-mine or not.

Background

• Landmines are a major problem in many parts of the world.

• They can be detected with a metal detector type device.

• The metal detect outputs a voltage that is proportional to the amount of metal in the ground.

• The measurement is effected by the height of the sensor above the ground and the amount of moisture in the soil.

The Dataset

We will be using a dataset that is described in

• Yilmaz, C., Kahraman, H. T., & Söyler, S. (2018). Passive mine detection and classification method based on hybrid model. IEEE Access, 6, 47870-47888

The data has been made available on the UCI Machine Learning Repository,

• https://archive.ics.uci.edu/dataset/763/land+mines-1

About the Features

338 samples of measurements from a metal detector. Each has 3 features

• Voltage (V): The output voltage of the magnetic sensor
• Height (H): The height of the magnetic sensor from the ground
• Soil Type (S): A value that corresponds to the amount of moisture in the soil

About the Labels

Each sample belongs to one of 5 classes:

• 1: No Landmine
• 2: Anti-Tank Mine
• 3: Anti-Personnel Mine
• 4: Booby trapped Anti-Tank Mine
• 5: M14 Anti-Personnel Mine

There are roughly equal numbers of samples in each class.

Classification or Detection?

• We will start out by treating this as a 5-class classification problem.

• We will then reconsider the problem as a detection problem, i.e., with just two classes: Mine or No Mine.

Obtaining the Jupyter Notebook

If you have cloned and pulled the module's GitHub repository then you should see,

materials/labs/
├── 060_introducing_scikit_learn.ipynb
|-- ... etc
├── data
│   ├── data/Mine_Dataset.xls
│   ├── ... etc

The lab is 060_introducing_scikit_learn.ipynb and it will need the data file data/Mine_Dataset.xls.

Or you can download the notebook and data via links on Blackboard.

Getting Help

• If you are stuck just raise a hand to ask for help.

• Feel free to discuss the lab with your neighbours.

• Re-read the Scikit-Learn tutorial notes

  • In the Git repo at materials/tutorials/060_Introducing_Scikit_Learn.ipynb
  • or online at https://uos-com-6018.github.io/COM6018

• Use the Scikit-Learn API documentation for reference. https://scikit-learn.org/stable/modules/classes.html