Drone Flight Logs Sequence Mining

Abstract

Data mining techniques in analyzing log data can discover a useful pattern which then is used to infer knowledge. Interesting patterns in log data can help the stakeholder to take action to diagnose a problem or improve the running system. Drone as one loT device, which consists of subsystems working together, also implements a logging mechanism. While a drone is flying, event-related logs are written into specific log files. These files contain precious information in case of incident happens to the drone. Assuming that the integrity of the log files is guaranteed, the investigator can find useful patterns and help conclude the incidents. To this end, this paper studies the sequence mining approach to discover some pre-defined incident-related events. As this is an initial study, the main contribution of this paper is the domain adaptation and modeling of the flight logs into a sequence database. After experimenting, we conclude that the modeling procedure is an essential step in conducting sequence mining. Frequency-oriented techniques are not suitable for small sequence databases, as the found patterns tend to have less critical events. Finally, two potential future directions are elaborated.