Activity Prediction using Wearable Sensor Data

This notebook explores and implements machine learning techniques for classifying physical activities using time-series data collected from wearable fitness devices. The assignment was originally completed during a university course and is now published for educational and portfolio purposes.

📊 Dataset

The dataset includes sensor readings from wearable fitness devices, stored in .dat files. It contains the following features:

• Timestamp (seconds)
• Activity ID
• Heart Rate (bpm)
• Electrodermal Activity (EDA)
• Temperature (Celsius)
• Accelerometer and Gyroscope (X, Y, Z axes)
• GPS (Latitude, Longitude)

Each row corresponds to a single timestamped sample for a specific user.

🔍 Project Overview

The notebook includes:

1. Data Loading and Preprocessing

   • Parsing .dat files into structured dataframes
   • Handling missing values and feature scaling

2. Exploratory Data Analysis (EDA)

   • Visualizing signal distributions and trends
   • Inspecting activity ID distributions

3. Feature Engineering

   • Creating time-windowed features
   • Combining data from different sensors

4. Modeling

   • Building and training a classification model using deep learning (CNN and/or LSTM)
   • Using Keras/TensorFlow for model definition and training
   • Evaluating performance using accuracy, confusion matrix, and F1 score

5. Results

   • Includes final training metrics, model accuracy, and plots from evaluation

📈 Results

The notebook includes all intermediate results, model training output, and final evaluation metrics.

🛠 Requirements

• Python 3.x
• Jupyter Notebook
• pandas, numpy, matplotlib, seaborn
• scikit-learn
• tensorflow or keras

To install dependencies:

pip install -r requirements.txt