Optimize Mobile App Testing Using Machine Learning to Improve User Experience

Abstract

Aims: The study delves into the machine learning (ML) paradigm shift in enhancing mobile application testing processes for higher accuracy, efficiency, and overall user experience, with a particular focus on Decision Tree and Random Forest models. Study Design: Experimental Research Design. Methodology: The research applies an experimental A/B testing framework using real-world datasets and cloud-based testing environments (e.g., Firebase Test Lab) to compare ML-driven and traditional testing approaches. Techniques include automated UI defect detection through convolutional neural networks, reinforcement learning for intelligent test case prioritization, natural language processing for extracting UX-related insights from user feedback, as well as a structured user survey involving 20 participants to evaluate perceived improvements in usability and stability. MATLAB R2024b was used for model development and evaluation. Results: Experimental results demonstrate that ML-based testing significantly outperforms traditional approaches, achieving 15–20% higher defect detection rates, 30–35% greater test coverage, and 40–50% faster execution times, alongside a notable reduction in false positives. Decision Tree and Random Forest models showed superior performance in identifying usability and performance issues. Additionally, the integration of ML into CI/CD pipelines facilitated faster bug resolution with minimal manual intervention. User survey results further confirmed improvements in user experience, with over 70% of respondents reporting enhanced application stability and responsiveness. Conclusion: Despite its promise, deploying ML in real-world testing presents challenges, including dataset bias, variability across device environments, and limited interpretability of some model decisions. To address these, the study recommends developing robust ML-based testing frameworks, ensuring access to representative and high-quality training data, and designing hybrid models that integrate supervised learning with unsupervised anomaly detection techniques.