Sequence Data Mining for Adverse Event Prediction and Action Recommendation

Abstract

Many real-life data mining applications use sequence data modeling in which data is represented as a sequence. A temporal sequence is finite ordered list of events (t1,e1), (t2,e2), ...,(tn,en) where ti represents time and ei represents the event taking place at time ti. ei takes place before ei+1 for 1≤ i ≤ n-1. This model can be used in data mining, called sequence data mining, to predict certain event that may take place at a specific time. Sequence data mining has a wide range of applications. This data mining technique can be used for prediction of adverse events and recommend proper actions to be taken as needed. In aviation safety, the future of a flight can be predicted as a sequence and proper action can be recommended to avoid dangerous situations that a flight may get into otherwise. In health care system, the future of a bacterial infection can be predicted and proper medicine can be prescribed for different situations to bring the patient’s illness to an end. In the marketing, customer shopping can be monitored and certain action can be taken, such as mailing coupons, to encourage the customer for further shopping of relevant products. In the real-life situations such as manufacturing, sensors’ data can be analyzed to control operations and predict dangerous situations and recommend proper actions to be taken. This paper discusses sequence representation, implementation, and its application for a number of different cases.