An agent-based framework for collaborative data mining optimization
Xiong Liu, Kaizhi Tang, John R. Buhrman, Huaining Cheng · 2010
Data mining meta-optimization aims to find an optimal data mining model which has the best performance (e.g., highest prediction accuracy) for a specific dataset. The optimization process usually involves evaluating a series of configurations of parameter values for many algorithms, which can be very time-consuming. We propose an agent-based framework to power the meta-optimization through collaboration of computing resources. This framework can evaluate the parameter settings for a list of algorithms in parallel via a multiagent system and therefore can reduce computational time. We have applied the framework to the construction of prediction models for human biomechanics data. The results show that the framework can significantly improve the accuracy of data mining models and the efficiency of data mining meta-optimization.