Mining Periodic Patterns For Moving Objects

With the ever-developing information technology and sensor, more various ways of obtainingare emerging now, which make it more convenient to get data. How to mine the patterns peopleare interested in from big data is becoming the research hotspot. Mining the periodic patterns inmoving objects provides theoretical foundations for reaching such moving objects as animalbehaviors and with the excavated periodic patterns, such reasonable decisions as traffic guidancecan be made. Compared with the conventional methods, there are many improved methods formining periodic patterns in recent years, but they are still not good enough. Worse still is that themethods for mining moving object’s periodic patterns are limited, so this thesis mainly researchesthe methods for mining periodic patterns in moving objects.To solve problems of combinatorial explosion and rare item as well as to satisfy the closureconditions and improve the efficiency, an algorithm of Multi-Constraint Closure Conditional Tree(MCCCT) is proposed. This method is able to set proper restrictive conditions for differentpatterns, which solves the combinatorial explosion and rare item problem and also improves themining efficiency. The attribute value is dynamically extracted according to each pattern’sfrequency of occurrence for getting pattern’s supports and period intervals, which enhances themining algorithm’s flexibility. To reduce the influence of uncertainties like noises, similarity-basedpattern matching algorithm is brought in to make the mining algorithm more robust. The methodproposed in the paper is verified and analyzed by using public moving object data. Each item’ssupport threshold and period interval threshold are gained based on their occurrence frequencies.The experimental results show that the method we put forward solves the problems that theconventional one cannot, for the method not only can extract the rare items, resist noises, but alsocan achieve periodic patterns with higher quality and satisfy the closure conditions.As redundancy exists in patterns extracted by the conventional method, the periodic patternsare not what users want. Therefore, a non-redundant period pattern extracting algorithm (NRPP) isput forward. The method also brings in multi-support constraint to improve sifting performance.With similarity factor being brought in, the similar patterns are trimmed so as to reduce theredundant patterns. By using public moving object data, the experimental results show that themethod we proposed can achieve more concise patterns with higher qualities than theconventional ones, in which there are no similar patterns. As a consequence, the proposed methodcan effectively reduce redundancy and make up conventional methods’ shortcomings.