Abstract: There are many automated vulnerability detection methods for Android applications. However, existing vulnerability detection solutions still rely on prior knowledge and lead to high false positive rates. To improve the existing vulnerability detection methods, a machine learning based method was proposed to identify the component exposure vulnerability of Android applications. Analyzing Android application software structure and component exposure vulnerability model, a new machine learning system was established to perform the Android vulnerability features extraction, data cleaning and vectorized operation. Utilizing manual analysis and verification, 1 000 Android APK sample sets were established. Through a large number of training, the system can detect the component exposure vulnerabilities automatically, achieving the accuracy up to 90%.