App Permission Classification dynamic Model(APCM)

Android dominates the smartphone operating system market, with 2.5 billion active users worldwide. The increasing dependability of smartphones poses user data as a crucial part of the smartphone’s security. Android publishes several apps daily, and tracking and restricting the apps is hard. The apps tend to mine the user data without the user’s knowledge. The user allows most permissions without reviewing and considering their future consequences. Hence, unwanted permissions can breach a user’s privacy and valuable data like personal information and banking details.The paper proposes a dynamic approach App Permission Classification dynamic Model (APCM), to track the app permission requested by the apps. The model uses a 2 million app dataset from Google Play Store to train a random forest-based model using DBSCAN clustering. The APCM creates a frequency map of the permissions based on the frequency of the permission request in a category. It uses DBSCAN clustering to rate the app permission as safe and unsafe and predicts using a random forest machine learning algorithm. The APCM uses PSO-BO optimisation over the random forest to enhance the results. The random forest with PSO-BO optimisation achieves an industry-leading accuracy of 87% in classifying an app’s permission as safe and unsafe. The APCM classifies the apps using the dataset and uses user preferences to alter the results and provide a better experience.