Human Activity Recognition Using Machine Learning

Project Overview

This project focuses on recognizing and classifying human activities based on sensor data collected from wearable devices. The goal is to accurately identify activities such as walking, sitting, standing, and more, using classical machine learning techniques.

Technologies & Tools

• Programming Language: Python
• Libraries: Pandas, NumPy, Scikit-learn, Matplotlib, Seaborn
• Environment: Jupyter Notebook

Dataset

The dataset consists of time-series accelerometer and gyroscope readings captured from multiple participants performing various activities. The data includes multiple features representing sensor signals, which are used to train and evaluate the models.

Methodology

• Data Preprocessing: Handling missing values, normalization, and feature extraction
• Model Training: Implemented classifiers including Random Forest and Support Vector Machine (SVM)
• Evaluation: Model performance assessed using accuracy scores and confusion matrices to analyze classification effectiveness

Results

• Achieved an accuracy of [96.67]% using Linear SVC on the test set
• Confusion matrix indicates strong classification performance across most activity classes
• Visualizations provide insight into data distribution and model predictions

📄 Capstone Report

This project was developed as a capstone project for the B.Tech program at SRM University-AP under the guidance of Dr. Pandu Sowkuntla.

📘 Click here to read the full technical report (PDF)

Summary of the Report

• Title: Human Activity Recognition Using Smartphones
• Team Members: Padala Saket Sai, Penubothu Gautham Sai Swaroop, Singamaneni Sriram, Kondavaradala Deepak Manidra
• Advisor: Dr. Pandu Sowkuntla
• Institution: SRM University AP, Computer Science & Engineering
• Submission Date: May 2024

Key Highlights

• Used smartphone sensor data (accelerometer & gyroscope) to classify 6 activities.
• Trained and compared multiple models: Logistic Regression, SVM (Linear & RBF), Decision Trees, and Random Forest.
• Achieved ~96% accuracy using SVM and Random Forest with hyperparameter tuning.
• Included data preprocessing, feature engineering (561 features), and t-SNE visualizations.
• Demonstrated applications in healthcare, activity monitoring, and smart devices.