Machine learning techniques for the determination of vehicle hijacking spots using twitter data
Loading...
Date
2024
Authors
Journal Title
Journal ISSN
Volume Title
Publisher
University of the Western Cape
Abstract
A vehicle hijacking is one of the leading crime-related incidents in South Africa. Each day, many travelers are caught unprepared due to a lack of knowledge regarding incident locations. This information is usually not easily accessible to the general public, and the currently available information is commonly released in large time increments, such as monthly or yearly reports. Therefore, an alternative approach to obtaining this data is needed. Social media provides an open-source alternative for data collection. One of the largest of these platforms is Twitter (newly named X.com). With Twitter, users share information regarding each aspect of their lives, such as notable daily incidents. It is also quite common for certified news outlets to inform users of current events as they occur. However, when dealing with Twitter data, the issue of relevant information is encountered. Due to the users’ free roam regarding topics and textual format, not all obtained data would be a relevant hijacking report. To remedy this, the employment of Machine Learning is observed. In nature, this is a textual classification problem. When dealing with such problems, there has been a large number of works employing supervised and unsupervised learning methods. For supervised learning approaches, an issue is a need for data to train a defined model, this removes the possibility of a true real-time approach. Unsupervised learning voids this requirement through the learning as occurring nature, however, it has commonly been found to have a reduction in performance. Therefore, variations of both methods are implemented in this work.
Description
Keywords
Machine learning techniques, Vehicle hijacking spots, Twitter, Crime-related incidents, South Africa