intelligence-engineering-CW2

Random Forest Activity Classification Pipeline

Project Description

This project implements a pipeline for activity classification using sensor data from wearable devices. The primary objective is to classify human activities such as Walking, Running, Lying Down, and Sitting based on features extracted from accelerometer sensors placed on different body parts. The project uses a Random Forest Classifier to achieve accurate activity recognition.

Features

• Preprocessing pipeline for handling missing values, mapping labels, and filtering activities.
• Feature selection using SelectKBest.
• Random Forest Classifier for activity prediction.
• Hyperparameter tuning using GridSearchCV.
• Model evaluation with metrics such as accuracy, classification report, and confusion matrix.

Libraries Used

The following libraries are used in this project:

• os
• pandas
• sklearn
  • train_test_split
  • RandomForestClassifier
  • classification_report, confusion_matrix, accuracy_score
  • GridSearchCV
  • SelectKBest, f_classif
• matplotlib
• seaborn

Dataset

The dataset contains sensor readings from wearable devices. Each data entry includes measurements such as back_x, back_y, back_z, thigh_x, thigh_y, and thigh_z and their corresponding activity labels.

Instructions to Set Up and Run the Project

Prerequisites

1. Python 3.8 or higher.
2. The required libraries listed in requirements.txt.

Setup Steps

1. Clone this repository:

   git clone https://github.com/Emem-studio/intelligence-engineering-CW2.git
   cd intelligence-engineering-CW2

2. Install the dependencies:

   pip install -r requirements.txt

3. Ensure your dataset files are in the appropriate directory (har70plus) as specified in the script.

4. Run the main script:

   python random_forest_pipeline.py

Output

• The script outputs accuracy, classification reports, and confusion matrices for the classification task.
• Best hyperparameters from GridSearchCV will also be displayed.

Dataset Details

• Features:
  • back_x, back_y, back_z
  • thigh_x, thigh_y, thigh_z
• Activities:
  • Walking
  • Running
  • Lying Down
  • Sitting
• Filtered out: Standing

Model Details

• Classifier: Random Forest
• Hyperparameter Tuning:
  • n_estimators
  • max_depth
  • max_features
• Balanced class weights to handle class imbalance.

Results

The final model achieves an accuracy of ~83.9% with the following performance metrics:

• Precision, Recall, and F1-Score for each activity.
• Confusion matrix heatmap visualization.

Automation

• This project integrates with GitHub for CI/CD and automation.
• Set up on Azure for scalable and efficient processing.