An interactive approach to multiobjective clustering of gene expression patterns

Area of Science:

• Computational Biology
• Data Mining
• Optimization

Background:

• Data clustering is often framed as a multiobjective optimization problem, requiring simultaneous optimization of multiple cluster validity indices.
• Existing cluster validity indices have limitations, performing variably across different datasets and failing to universally capture optimal clustering structures.
• Selecting the optimal set of validity indices for simultaneous optimization is crucial for achieving robust clustering outcomes.

Purpose of the Study:

• To propose a novel interactive genetic algorithm-based multiobjective approach for data clustering.
• To simultaneously determine the optimal clustering solution and evolve the most effective set of cluster validity indices.
• To incorporate human decision-maker input interactively for adaptive learning and improved clustering performance.

Main Methods:

• Development of an interactive genetic algorithm designed for multiobjective optimization in data clustering.
• The algorithm adaptively learns from human decision-maker (DM) input during execution to refine the set of validity indices.
• Application of the proposed method to cluster real-life benchmark gene expression datasets.

Main Results:

• The proposed interactive multiobjective approach successfully identified effective sets of validity indices and achieved optimal clustering solutions.
• Performance comparison demonstrated that the novel method significantly outperforms existing clustering algorithms on the tested gene expression datasets.
• The adaptive learning mechanism, guided by DM input, proved effective in enhancing clustering accuracy and robustness.

Conclusions:

• The interactive genetic algorithm-based multiobjective approach offers a powerful and adaptive solution for data clustering.
• This method effectively addresses the challenge of selecting appropriate cluster validity indices, leading to superior clustering performance.
• The findings highlight the potential of interactive optimization techniques in computational biology and data mining applications.