Identification of Bot Calls

Challenge

Calls worldwide are routed to their respective countries through so-called "routes," which can be either direct or indirect. Unlike Western countries, emerging markets often impose high fees for officially connecting calls to their countries. In response, fully or partially automated bot calls are made to Europe to identify routes not officially registered. Our client, a routing services operator, faced the challenge of automatically identifying bot calls to blacklist them effectively.

Approach

Based on historical call detail records, an aggregated dataset was created, compressing various features at the phone number level. These features include the number of connections, call behavior in terms of time of day and seasonality, the entropy of called numbers, and many more. These characteristics formed the basis for training a machine learning model capable of identifying bot calls with high accuracy.

Result

Based on the estimated machine learning model, a solution was implemented to regularly score callers, with results visualized interactively in a dashboard. This allows client-side employees to conduct further detailed analyses, helping them better understand the bot call systematics and further develop appropriate defense mechanisms.