Mining Transactional Data To Produce Extended Association Rules Using Collaborative Apriori, Fsa-Red And M5p Predictive Algorithm As A Basis Of Business Actions

Feri Sulianta,Laksana Eka Angga,Thee Houw Liong

2024-03-07

Abstract:There are large amounts of transactional data which showed consumer shopping cart at a store that sells more than 150 types of products. In this case, the company is utilizing these data in making business action. In previous studies, the data that has a lot of attributes and record data reduction algorithms handled by the FSA Red (Feature Selection for Association Rules)are then mined using Apriori algorithm. The resulting association rules have high levels of accuracy and excellent test results, which rely more than 90%. In this study, the association rules generated in previous research will be updated by using prediction algorithms M5P, so that the association rules can be used within a period of several months in the future. Furthermore, some data mining technique such as: clustering and time series pattern will be implemented to examine the truth and extend the validity of association rules which were built. It can be concluded that the association rules were established after will generate strong association rules with confidence equal or higher than 70% and the rules established truth can be seen from the time series pattern on each group of goods which are then used as the basis of business actions.

Databases

What problem does this paper attempt to address?

This paper discusses the use of big data mining technology to generate updatable association rules to support business decisions. The study mentions that a large amount of transaction data reflects consumers' shopping behavior in stores that sell a variety of products. The company hopes to use this data for market basket analysis to understand consumers' purchasing patterns and formulate business strategies.

Based on previous research, the paper proposes a new method. Firstly, a feature selection algorithm called FSA-Red is used to reduce the dimensionality of a large number of attributes and record data. Then, the Apriori algorithm is applied to mine high-precision association rules. The accuracy of these rules exceeds 90%. In order to ensure the reliability and practicality of the rules for the future, the researchers introduce a prediction algorithm called M5P to update the association rules. In addition, clustering and time series pattern analysis are used to verify and expand the effectiveness of the association rules.

With this method, strong association rules with a confidence level equal to or higher than 70% can be obtained, and the patterns of each product group can be observed through time series analysis, serving as the basis for business actions. The results show that the established association rules can generate predictions with high accuracy, particularly in the next few days. However, the accuracy may decrease over time, so it is recommended to limit the use of rules with an accuracy higher than 70%.

The paper also proposes several future research directions, including using different data reduction algorithms, considering support values, exploring other prediction algorithms, improving clustering methods, and using more data for training and testing to obtain better results. Overall, the research aims to improve the long-term stability and commercial value of the association rules extracted from transaction data.