AI-Powered Smartphone Context and High-Utility Itemset Mining for Enhanced App Testing and Personalization

Internet of Things (IoT) devices have become increasingly essential. User experience and functionality rely heavily on context events such as screen orientation, wireless connectivity, media events, and battery events. Volatility and unpredictability of these events pose challenges to app development and testing. To address this, we use a novel application of SPMF library algorithms, including Top-K Quantitative High Utility Itemset Miner (TKQ), Fast High Utility Quantitative Itemset Miner (FHUQI-Miner), and Fast Correlated High-Utility Itemset Miner (FCHM) with two different correlation metrics. We train the machine learning techniques (decision tree regression (DTR) and bagging on generated rules on itemsets) from sequences of context events. This includes 78 unique context event streams from real world users over periods of 15 days, 30 days, and 60 days. We identify high-utility itemsets (HUIs) of context events, which developers can incorporate into testing. Results show that the HUIs mining algorithms result in high predictive accuracy of greater than 90% on testing data which makes it suitable for helping to improve context-aware software testing processes. Indeed, HUIs help predict user behavior, personalize user experience, detect anomalies, and realize efficient and reliable mobile apps through context-aware testing.