ABSTRACT
In recent years, mobile applications have made crucial contributions to day-to-day lives. At present, these applications are being utilized by almost all individuals. Including social media applications, messaging, browsers, etc., on Google playstore, users can select millions of mobile applications that correspond to their devices. Therefore, detecting the individual’s application usage pattern is vital for the current data marketers to comprehend the demand and needs of the users. Concurrently, ML (Machine Learning) based algorithms have gained significance in this area due to their innate capability to identify patterns easily. Existing research has endeavored to use different ML-based algorithms to determine application usage patterns. However, such studies have been ineffective in overfitting and low prediction rate. This study proposes MGA (Modified Genetic Algorithm) for selecting relevant features to attain high pattern prediction. As the conventional GA can get trapped into a local optimum with an objective function and similarity in parameter inversion, an elitist and multi-objective optimization approach is introduced to resolve the inverse issue.
Further, the research proposes MK-MA (Modified K-Means Algorithm) based on the Dynamic Density Updated Isolated Forest. In this process, the density updation is considered to solve the challenges of unstable prediction outcomes caused by the random partition of dataset features in Isolation Forest. The overall performance is comparatively evaluated by performance metrics that expose the efficacy of the proposed system in pattern recognition.
