Automatic Annotation of Unlabeled Data from Smartphone-Based Motion and Location Sensors.

Area of Science:

• Data Science
• Machine Learning
• Mobile Crowd Sensing

Background:

• Manual annotation of sensor data is challenging and reduces user experience.
• Unsupervised learning, specifically clustering, can automate data annotation but requires robust cluster validation.
• Detecting sensitive location information from mobile crowd sensing data is crucial for user privacy.

Purpose of the Study:

• To automatically annotate unlabeled sensor data for mobile crowd sensing.
• To propose a novel cluster validation index for the k-means algorithm.
• To evaluate the performance of supervised classifiers using automatically generated cluster labels.

Main Methods:

• Utilized the k-means clustering algorithm for unsupervised annotation of sensor data.
• Developed a new cluster validation index based on Multiple Pair-Frequency.
• Trained and evaluated Support Vector Machine, K-Nearest Neighbor, and Naïve Bayes classifiers on annotated data.

Main Results:

• The k-means algorithm with the proposed validation index effectively annotated sensor data.
• Support Vector Machine and K-Nearest Neighbor classifiers achieved very high accuracy in classifying sensitive and non-sensitive data.
• Naïve Bayes classifier demonstrated fairly high accuracy.

Conclusions:

• A hybridized approach combining unsupervised (k-means clustering) and supervised (classification) machine learning effectively annotates and classifies sensor data.
• The proposed Multiple Pair-Frequency based cluster validation index enhances the reliability of automatic annotation.
• This method significantly reduces manual effort and improves the detection of sensitive location information in mobile crowd sensing.