Ships' Association Rule Mining Based on Adaptive Support Threshold and Arcsin Interestingness

Abstract

There exists two shortcomings in the process of mining association rules from AIS data of ships by traditional method based on support and confidence: First, the types and quantities of ships involved in the AIS database are numerous. The period of broadcasting AIS may be different from each ship, and the number of generated transactions may also be different. if the count of the transaction containing one itemset is less but the relationship between items in it is very close, the criterion with high support threshold may overlook it directly. However, this kind of itemsets is significantly more representative and valuable. Conversely, if the itemset occurs more frequently but the relationship between items of it is not close, the traditional criterion method will also discriminate it as a frequent itemset. But this kind of itemset is of little significance and credibility; Second, The association rule mining method based on support and confidence may obtain the rules that users are not interested in or even inaccurate. To solve these problems more effectively, the ships' association rule mining algorithm based on Adaptive Support Threshold and Arcsin Interestingness (ASM-AI Algorithm) is presented in this paper. On the one hand, each candidate itemset will have the different adaptive support threshold to solve the two misjudgment problems, and the paper demonstrates that the new discrimination constraint has the antimonotonicity. On the other hand, the concept of Minimum Lift Function is introduced as an index to evaluate the credibility of association rule discrimination method and the Arcsin Interestingness is proposed as a new interestingness measure. It is proved by mathematical deduction that compared with some existing research results, this method can improve the reliability of association rules. Furthermore, the concepts of Lift are proposed to evaluate the performance of the algorithm's result. Finally, the real AIS data is used to test algorithm performance and the result shows that the new discrimination criteria of the ASM-AI Algorithm is more stringent, the more reliable association rules can be found out and the lift of association rules obtained by ASM-AI Algorithm is generally larger than the existing research's results.