A Cooperative Multi-agent Data Mining Model and Its Application to Medical Data on Diabetes

Abstract

We present CoLe, a model for cooperative agents for mining knowledge from heterogeneous data. CoLe allows for the cooperation of different mining agents and the combination of the mined knowledge into knowledge structures that no individual mining agent can produce alone. CoLe organizes the work in rounds so that knowledge discovered by one mining agent can help others in the next round. We implemented a multi-agent system based on CoLe for mining diabetes data, including an agent using a genetic algorithm for mining event sequences, an agent with improvements to the PART algorithm for our problem and a combination agent with methods to produce hybrid rules containing conjunctive and sequence conditions. In our experiments, the CoLe-based system outperformed the individual mining algorithms, with better rules and more rules of a certain quality. From the medical perspective, our system confirmed hypertension has a tight relation to diabetes, and it also suggested connections new to medical doctors.

CoLe: A Cooperative Data Mining Approach and Its Application to Early Diabetes Detection

“…Ag CBN also controls how D i 's are created for the next iteration. [3] The most important aspect of CoLe is the cooperation of miners, coordinated by Ag CBN : it instructs how D i 's are generated from D; it receives results K i from the miners; the final result K is produced by Ag CBN . Ag CBN also synchronizes the iterative mining process.…”

A Multi-Agent Approach to Data Mining Processes: Applications to Health Care

“…Today's modern health institutions automatically collect structured data relating to all aspects of care such as diagnosis, medication, test results and radiological imaging data (Jensen et al, 2012). Legacy systems, health monitoring devices, applications for disease management and fitness tracking (Szewczyk, 2016) and medical records data sets for research studies (Gao et al, 2005a) are sources that produce large volumes of data in regular basis.…”

“…Although there are already data mining solutions to deal with specific domains (Gao et al, 2005a;Palaniappan & Awang, 2008;Gorodetsky et al, 2003;Mitkas et al, 2003;Ralha & Sarmento, 2012;Gonzalez-Sanchez et al, 2011), these approaches do not explicitly present how to create a flexible architecture that take advantage of software agents to manage and perform mining and classification of data for different domains. Software agent technology brings several benefits to data mining, such as autonomy, reactivity, collaboration, distributed processing, among others (Cao et al, 2009).…”

“… Some solutions for data mining and classification do not provide mechanisms to be extended to other domains or at least evidence that they were actually extended to other domains (Gao et al, 2005a;Palaniappan & Awang, 2003;Ralha & Sarmento, 2012;Gonzalez-Sanchez et al, 2011);…”

“… Software agents allow the execution of parallel mining, which brings advantages such as the generation of several models at same time; nonetheless the use of ensemble techniques to make profit of combining those models (Di Stefano & Menzies, 2002;Woźniak et al, 2014;Alpaydin,1998) to produce new ones is almost absent (Gao et al, 2005a;Gorodetsky et al, 2003;Mitkas et al, 2003;Ralha & Sarmento, 2012;Gonzalez-Sanchez et al, 2011) as is any comparison between models;…”