Crank Call Detection Models Based on Call Data of Telephone Subscribers

Abstract

Data explosion makes our life much more convenient than ever. For example, people always get timely news from Apps, and get interesting recommendations when shopping online. However, these conveniences come by probing people’s private data, such as account information, search history, and chat records. Thus, if the management of data is not as secure as it supposes to be, many problems, e.g., crank calls, will arise and cause big troubles in people’s lives. Although many security Apps have been developed to help mobile phone users stay away from crank calls, the users need to update Apps frequently due to their user-labels-based blacklist. These feedback-based systems depend on a large number of user actions that have significant lag. In this paper, we propose two crank call detection models to help operators mark crank callers based on the call data of telephone subscribers, in which the features are organized by different ways. One is a composited classification model, and the other is a convolutional neural network model based on the time-sliced data. The experiments show that the model applying convolutional neural network gets better accuracies in detecting crank calls.