International Research Journal of Engineering and Technology (IRJET) e-ISSN: 2395-0056 Volume: 11 Issue: 07 | July 2024
www.irjet.net
p-ISSN: 2395-0072
A Comparative Study on Human Activity Recognition Using Smartphone Dataset through Machine Learning Approach Yashraj Mishra1, Ankita Jaiswal2, Dr. Goldi Soni3 1Student, Amity University Chhattisgarh 2Student, Amity University Chhattisgarh 3Assistant Professor, Amity University Chhattisgarh
---------------------------------------------------***----------------------------------------------------------
Abstract - Human Activity Recognition (HAR) using
activities: walking, walking upstairs, walking downstairs, sitting, standing, and lying down.
smartphone data leverages built-in sensors to detect and classify users’ activities. The proliferation of smartphones with various sensors has opened new avenues for recognizing human activities, with significant noteworthy applications in remote healthcare and activity tracking for the elderly or disabled, and fitness monitoring. Recent studies focus on applications that identify daily track activities and calculate calories burned in real-time. These applications capture labeled triaxial acceleration data from the smartphone’s accelerometer, which is then preprocessed and analyzed. Features are extracted using various methods, and machine learning algorithms classify the activities. The most effective models are integrated into smartphone applications, enabling real-time activity recognition and health monitoring. This project contributes to developing assistive technologies that improve quality of life and promote a healthier lifestyle through classification techniques. Six types of human activities: standing, sitting, lying down, walking, walking upstairs, and walking downstairs —were identified using data from the University of California Machine Learning Repository. Data from the Samsung Galaxy S II’s gyroscope and accelerometer were divided into training and testing sets in a 70:30 ratio. Principal Component Analysis reduced dimensionality, and machine learning methods such as Random Forest, Support Vector Machine, Artificial Neural Network, and K-Nearest Neighbor classified activities. Random simulation and confusion matrices compared the accuracy and performance of various models.
1.1 Background The smartphone's integrated accelerometer and gyroscope recorded 3-axial angular velocity and 3-axial linear acceleration at a steady rate of 50Hz. These recordings were essential for capturing the dynamic movements associated with each activity. To ensure accuracy in labelling the data, the trials were captured on camera, enabling manual verification of the activities performed. The generated dataset was divided randomly into two subsets: training data consisting of 70% of the volunteers and test data from the remaining 30%. This division ensured that the machine learning models could be trained and validated effectively.
1.1.1 Pre-processing and Feature Extraction Raw accelerometer and gyroscope signals underwent pre-processing to enhance data quality. Noise filters removed irrelevant data, and signals were sampled using 2.56-second sliding windows with 50% overlap, yielding 128 readings per window. A Butterworth low-pass filter with a 0.3 Hz cut off was applied to distinguish between gravitational and body motion components. Features were extracted from each window, including time and frequency domain measures like mean, standard deviation, energy, and entropy, to support accurate machine learning classification.
Key words: Exploratory Data Analysis, Machine Learning Algorithms, Data Modelling, Hyperparameter and cross validation.
1.1.2 Dataset and Machine Learning Application The UCI Machine Learning Repository, in collaboration with Team Kaggle, provided a dataset consisting of two main files: a training set with 7,352 rows and 564 columns, and a test set with 2,947 rows and 564 columns. Each row represents a sliding window of sensor data, and each column corresponds to a specific feature derived from the raw signals. The goal was to develop models to learn from the training data and accurately predict activity labels on the test data. Various algorithms, including decision trees, support vector
1.INTRODUCTION Human Activity Recognition (HAR) using smartphone data is an emerging field that leverages machine learning algorithms to classify various physical activities based on information collected from a smartphone's sensors. This study involved 28 individuals aged between 19 and 48, who participated in a series of experiments. Each participant had a smartphone (Samsung Galaxy S II) attached to their waist while carrying out six different
© 2024, IRJET
|
Impact Factor value: 8.226
|
ISO 9001:2008 Certified Journal
|
Page 813