Association rule learning

About: Association rule learning is a research topic. Over the lifetime, 15194 publications have been published within this topic receiving 362099 citations. The topic is also known as: association rule mining.

Introduction to Supervised Descriptive Pattern Mining

Abstract: This chapter introduces the supervised descriptive pattern mining task to the reader, providing him/her with the concept of patterns as well as presenting a description of the type of patterns usually found in literature. Patterns on advanced data types are also defined, denoting the usefulness of sequential and spatiotemporal patterns, patterns on graphs, high utility patterns, uncertain patterns, along with patterns defined on multiple-instance domains. The utility of the supervised descriptive pattern mining task is analysed and its main subtasks are formally described, including contrast sets, emerging patterns, subgroup discovery, class association rules, exceptional models, among others. Finally, the importance of analysing the computational complexity in the pattern mining field is also considered, examining different ways of reducing this complexity.

A fast algorithm for maximum frequent itemsets based on the user' interest using FP-matrix

Abstract: Mining maximum frequent item sets is crucial for mining association rules. This paper proposes a novel algorithm, Interest Frequent Pattern Matrix(IFPM), which is based on user's interests, and illustrates the algorithm's execution process. IFPM preprocesses and filters the transaction database according to the level of data item and user's interests, making the handling data reduce an order of magnitude. And then scans the filtered database to create a FP-Matrix, searches the FP-Matrix by top-down depth-first, which can produce maximum frequent item sets(MFI), frequent item sets(FI)and Closed frequent item set(CFI) by vector operation, thus greatly improves the algorithm's efficiency.

Combined mining approach to generate patterns for complex data

Abstract: In Data mining applications, which often involve complex data like multiple heterogeneous data sources, user preferences, decision-making actions and business impacts etc., the complete useful information cannot be obtained by using single data mining method in the form of informative patterns as that would consume more time and space, if and only if it is possible to join large relevant data sources for discovering patterns consisting of various aspects of useful information. We consider combined mining as an approach for mining informative patterns from multiple data-sources or multiple-features or by multiple-methods as per the requirements. In combined mining approach, we applied Lossy-counting algorithm on each data-source to get the frequent data item-sets and then get the combined association rules. In multi-feature combined mining approach, we obtained pair patterns and cluster patterns and then generate incremental pair patterns and incremental cluster patterns, which cannot be directly generated by the existing methods. In multi-method combined mining approach, we combine FP-growth and Bayesian Belief Network to make a classifier to get more informative knowledge.

Exploiting efficient parallelism for mining rules in time series data

Abstract: Mining interesting rules from time series data has earned a lot of attention to the data mining community recently. It is quite useful to extract important patterns from time series data to understand how the current and the past values of patterns in the multivariate time series data are related to the future. These relations can basically be expressed as rules. Mining these interesting rules among patterns is time consuming and expensive in multi-stream data. Incorporating parallel processing techniques is helpful to solve the problem. In this paper, we present a parallel algorithm based on a lattice theoretic approach to find out the rules among patterns that sustain sequential nature in the multi-stream data of time series. The human motion data considered as multi-stream multidimensional data used as data set for this purpose is transformed into sequences of symbols of lower dimension due to its complex nature. Then the proposed algorithm is implemented on a Distributed Shared Memory (DSM) multiprocessors system. The experimental results justify the efficiency of finding rules from the sequences of the patterns for time series data by achieving significant speed up comparing with the previous reported algorithm.

Knowledge Discovery of Selection Rules for Acupuncture Points in Respiratory Diseases Therapy Based on Partial-Ordered Structure Diagrams

Abstract: This paper presents a knowledge discovery method of selection rules for acupuncture points in respiratory diseases therapy based on the theory of Structural Partial-Ordered Attribute Diagram and association rule mining. First, we briefly introduced the basic definitions of Structural Partial-Ordered Attribute Diagram and association rule mining theory. Then, we transformed the data of a Traditional Chinese Medicine treatise into formal context and transaction database. Finally, we explained knowledge discovery process by analyzing the formal context of respiratory diseases. It was concluded that the method proposed in this paper works well in discovering new knowledge from medical treatises and clinical cases of acupuncture treatment. The method provided a scientific and advanced technological means for the heritage of Traditional Chinese Medicine.