Context-Sensitive Systems (CSS) - Lab Tasks

Sensor data collection, analysis, and machine learning for human activity recognition.

Overview

This repository contains four lab tasks focused on sensor-based activity recognition:

1. Task 1: Web-based sensor data collection (orientation, accelerometer, gyroscope)
2. Task 2: Time series feature extraction with sliding windows
3. Task 3: Activity classification with 9 ML algorithms and LOSO cross-validation
4. Task 4: Hyperparameter optimization using Bayesian methods and Optuna

Quick Start

# Clone the repository
git clone https://github.com/LiliVanilli/css.git
cd css

# Navigate to specific task
cd task1  # or task2, task3, task4

Task Summaries

Task 1: Sensor Data Collection

• Goal: Build web app to collect device sensor data
• Tech: HTML5, JavaScript, Device APIs, EdgeML
• Live Demo: https://lilivanilli.github.io/css/
• Features: Real-time sensor display, CSV export, mobile support

Task 2: Feature Extraction

• Goal: Extract statistical features from time series data
• Tech: Python, Jupyter, Pandas, EdgeML API
• Method: Sliding windows with mean/variance calculation
• Output: Processed features ready for ML

Task 3: Activity Classification

• Goal: Compare ML algorithms for activity recognition
• Dataset: 144 samples, 8 subjects, 4 activities
• Methods: 9 classifiers with LOSO & 10-Fold CV
• Best Result: Random Forest F1=0.518 (LOSO), F1=0.748 (10-Fold)

Task 4: Hyperparameter Optimization

• Goal: Optimize Random Forest with Bayesian methods
• Methods: Custom GP, Optuna (TPE, CMA-ES, Random)
• Features: Dynamic feature extraction (28 features from heterogeneous sensors)
• Best Result: F1=0.77 (+48.6% vs baseline), 30 efficient iterations

Key Results

Task	Method	Result	Improvement
Task 1	Data Collection	Web app with 3 sensors	-
Task 2	Feature Extraction	Sliding windows + stats	-
Task 3	Random Forest (LOSO)	F1 = 0.518	Baseline
Task 4	Bayesian Optimization	F1 = 0.770	+48.6%

Technologies

• Frontend: HTML5, CSS3, JavaScript (ES6+)
• Backend: Python 3.10+, Jupyter Notebooks
• ML Libraries: scikit-learn, Optuna
• Data: Pandas, NumPy
• Visualization: Matplotlib, Seaborn, Plotly
• Deployment: GitHub Pages (HTTPS)

Repository Structure

css/
├── task1/                  # Web sensor data collection
│   ├── index.html
│   ├── script.js
│   └── README.md
├── task2/                  # Feature extraction
│   ├── analysis.ipynb
│   └── README.md
├── task3/                  # ML classification
│   ├── classification_clean.ipynb
│   └── README.md
├── task4/                  # Hyperparameter optimization
│   ├── hyperparameter_optimization.ipynb
│   └── README.md
└── README.md              # This file

Installation

Each task has its own dependencies. Navigate to the task folder and follow its README.

For Python tasks (2, 3, 4):

cd task3  # or task2, task4
pip install -r requirements.txt
jupyter notebook

For web task (1):

cd task1
npm install
open index.html

Author

LiliVanilli

License

MIT