A Heuristic Approach for Fast Mining Association Rules in Transportation System

Abstract

This paper proposes a heuristic algorithm for fast mining association rules by multidimensional scaling (MDS). It takes the similarity measurements as the MDS proximities and develops a practical MDS model to generate decentralized configuration of points that represent the stops on vehicle routes. This algorithm extends the SMACOF algorithm by the steps of grouping and join. The experiments show that the novel algorithm has much higher efficiency than the Apriori algorithm especially when mining association rules of long patterns in transportation system.

A survey on association rules mining using heuristics

“…First, we evaluated algorithms that were originally compared with Apriori. The results illustrate that NICGAR (Martín et al, 2016), MONPNAR (Martín et al, 2014), G3PARM (Luna et al, 2012), MDS-H (Hong & Bian, 2008), Ant-ARM (He & Hui, 2009), and SRmining (Hong-yun et al, 2008) are the fastest heuristic ARM algorithms compared to Apriori. At that point, we compared approaches that compared themselves with other heuristic approaches.…”

“…According to Figure and the above information, NICGAR (Martín et al, ) is 575 times, MONPNAR (Martín et al, ) is 533 times, G3PARM (Luna et al, ) is 23 times, and MDS‐H (Hong & Bian, ) is about nine times faster than Apriori. After that, ASC (Kuo & Shih, ) is approximately four times faster than Apriori.…”

“…Another heuristic-based ARM approach is MDS-H, which have been presented in Hong and Bian (2008). It is based on the multidimensional scaling (MDS) (Hong & Bian, 2008) approach and tries to improve Apriori in terms of performance. MDS-H consists of two steps: grouping and joining.…”

“…It may increase some other performance metrics like memory usage. Prasanna, Seetha, & Kumar, 2014 noted that the memory overhead of parallel processing could have a negative impact on system performance (Hong & Bian, 2008). One of the solutions that they proposed is to employ a memory aware resource allocation policy.…”

“…After all, according to these comparisons, NICGAR (Martín et al, 2016), MONPNAR (Martín et al, 2014), G3PARM (Luna et al, 2012), MDS-H (Hong & Bian, 2008), SRmining (Hong-yun et al, 2008), PMES (Djenouri et al, 2014), Ant-ARM (He & Hui, 2009), and Kuoa et al (2011) have the potential to be the fastest heuristic ARM algorithms. However, there is a big gap in this area and all the heuristic approaches should compare themselves with each other and instead of comparing themselves with Apriori, they should try to establish comparisons with more up-to-date and faster ARM algorithms.…”