ML Pipeline for Surface Classification Based on IMU data

Team

Mehdi Raza Khorasani & Ozan Pali

Table of Contents

• Introduction
• Dataset
• Installation
• Usage
• Pipeline Overview
• Data Exploration
• Data Preprocessing
• Data Visualization
• Class Imbalance & Evaluation Metrics
• Machine Learning Models
• Results
• Contributing
• License

Introduction

This project aims to develop a machine learning pipeline for classifying different surface types using IMU sensor data. The dataset consists of non-linear and imbalanced features, making classification a challenging task. The project applies multiple classifiers and evaluates their performance using appropriate metrics.

Dataset

The dataset contains input features (X) and labels (y). The data is highly non-linear and exhibits class imbalance. The primary goal is to process this data effectively and build robust classification models.

Installation

To set up the project, you can use locally hosted jupyter backend or colab/kaggle.

Pipeline Overview

The pipeline consists of the following steps:

• Data exploration and preprocessing
• Splitting the dataset into training and testing subsets
• Visualizing the data
• Training multiple classifiers (LGBM, KNN, Gradient Boosting, Random Forest)
• Evaluating performance using accuracy, precision, recall, and F1-score

Data Exploration

The dataset was analyzed for missing values and inconsistencies. The conclusion was that the data is not sparse and is ready for further processing.

Data Preprocessing

• Scaling Features (X): The features were standardized to have zero mean and unit variance, ensuring fair distance-based comparisons.
• Label Encoding (y): Labels were converted into numerical format to be compatible with machine learning algorithms.
• Train/Test Split: The dataset was divided into 80% training and 20% testing sets.

Data Visualization

• Class Distribution: The dataset exhibits class imbalance, where certain surface types have significantly fewer samples.
• Feature Distribution: PCA and t-SNE were applied to visualize feature separability, but the data remains highly non-linear and difficult to separate.

Class Imbalance & Evaluation Metrics

• Due to class imbalance, accuracy alone is not a reliable metric.
• Instead, precision, recall, and F1-score were used for model evaluation.
• Metric Definitions:
  • Precision: Measures how many predicted positive cases are actually positive.
  • Recall: Measures how many actual positive cases were correctly identified.
  • F1-Score: A balanced measure that considers both precision and recall.
  • Confusion Matrix: Provides a breakdown of correct and incorrect predictions.

Machine Learning Models

The following models were implemented:

• Random Forest: An ensemble method combining multiple decision trees for robust classification.
• Gradient Boosting: Sequentially improves weak learners by minimizing residual errors.
• LightGBM (LGBM): A faster and more efficient variant of gradient boosting.
• K-Nearest Neighbors (KNN): A distance-based algorithm relying on the majority class of neighbors.

Results

• Best Performing Models: Random Forest and LGBM
• Performance Metrics:
  • Accuracy: ~97%
  • Precision: ~97%
  • Recall: ~97%
  • F1-Score: ~97%

Contributing

Contributions are welcome! Please follow these steps:

1. Fork the repository
2. Create a new branch (git checkout -b feature-branch)
3. Commit your changes (git commit -m 'Add feature')
4. Push to the branch (git push origin feature-branch)
5. Open a Pull Request