Research On Personal Behavior Prediction Based On Behavior Entropy

Behavior prediction which was referred to as divination in the old days has now gradually become a socially valuable research topic today. In the recent years, behavior prediction has showed up in most research fields. This is true especially in the emerging field of big data processing and data mining which provides powerful support for personal behavior prediction. The predicton theory has finally stepped up and is now ready for scientific development.In this paper, we try to construct a computing model based on entropy theory to quantify personal behavior, with the help of analysis of behavior characteristics. To reach that goal, the first thing we have to do is to figure out those characteristic attributes which determined the behaviors. Therefore, we firstly discussed the characteristic attributes on several classifications such as intuitive display, value range and correlativity. Note that a characteristic attribute might lead to two kinds of results; we analyzed the relativity of characteristic attributes.After that, we proposed a conception named consequence weight to estimate the importance of attributes. Different attributes can have different numerical influence value whether the influence is positive or negative. In most cases the greater the influence value the greater the consequence. Hence, the thesis discusses the best way to calculate the weight of characteristic attributes in detail. That lays solid foundation for the computation of personal behavior entropy.With the definitions of characteristic attributes proposed, we proceed to evaluate the behavior trends using the behavior entropy theory which we put forward before. The behavior entropy theory draws on the experiences of information entropy and thermodynamic entropy. For this reason, behavior entropy builds direct relationships between macroscopic behavior and microscopic attributes.For the experiment, we analyzed the relationships among behavior attributes and characteristic attributes related to target behavior. The datasets we used came from the American Community Survey and the National Archive of Criminal Justice Data. In conclusion, we chose a famous person who is involved in social events and simulated his past behavior entropy. The experiments indicate that the theory can be used to predict the behavior of an individual based on behavior entropy.