|
|
 |
| |
|
|
|
MACT: A Novel Framework for Automated Mobile Application Testing Using Machine Learning |
|
|
|
PP: 1079-1092 |
|
|
|
doi:10.18576/amis/190509
|
|
|
|
Author(s) |
|
|
|
Moheb R. Girgis,
Alaa M. Zaki,
Enas Elgeldawi,
Mohamed M. Abdallah,
Ali A. Ahmed,
|
|
|
|
Abstract |
|
|
| The rapid expansion of mobile applications has amplified challenges in software testing, with inadequate testing responsible for 42% of application failures. Traditional testing methods are often time-consuming, resource-intensive, and hindered by issues such as GUI element identification and cross-device compatibility. This paper introduces MACT (Mobile Application Classification and Testing), a novel framework for automated mobile application testing that integrates machine learning-based activity classification with predefined, reusable test case execution. The framework automatically identifies and categorizes Android application screens (e.g., login, settings) and applies tailored test cases for each screen type, eliminating the need for custom test scripts and reducing maintenance overhead. Leveraging the MASC dataset, developed in our previous work, which comprises 7,065 screens from over 3,400 apps, the framework employs the Gradient Boosting model in activity classification that achieves 93.48% accuracy. Empirical evaluations demonstrate that MACT detects all planted bugs with 100% accuracy, significantly outperforming traditional tools like the Android Application Monkey. The framework reduces test script size by 87% and testing duration by 89% compared to manual methods like Android Espresso, providing a scalable, efficient, and resilient approach to mobile application testing. The modular design of MACT ensures seamless adaptation to various applications, addressing key limitations in GUI modeling and automated testing. This research makes a significant contribution to mobile application testing by reducing manual effort, increasing test coverage, and boosting reliability and efficiency.
|
|
|
|
|
|
|
|
