Classification of Microarray Gene Expression Data Using an Infiltration Tactics Optimization (ITO) Algorithm.

Area of Science:

• Machine Learning
• Optimization Algorithms
• Computational Science

Background:

• Existing feature selection and classification techniques present trade-offs between speed and accuracy.
• Parameter-free algorithms are fast but risk local optima, while parameter-based methods require tuning for accuracy but not necessarily reliability.
• Achieving generalized optimization for high accuracy and reliability remains an open research challenge.

Purpose of the Study:

• To present a novel warzone-inspired optimization algorithm, Infiltration Tactics Optimization (ITO), for creating high-accuracy-high-reliability (HAHR) binary classifiers.
• To combine the strengths of parameter-free and parameter-based classification methods.
• To address data scarcity issues and enhance model reliability.

Main Methods:

• The proposed ITO algorithm employs a two-phase approach: Lightweight Infantry Group (LIG) for rapid convergence to non-local maxima, and Followup Team (FT) for advanced tuning to boost performance.
• Each component ('soldier') within the ITO framework utilizes independently chosen subset selection, pre-processing, validation, and classification methods.
• Heterogeneous ensembles of successful 'soldiers' are combined for optimal results.

Main Results:

• The LIG phase achieves comparable results with 70-88% accuracy.
• The FT phase enhances baseline performance, yielding accuracies between 75-99%.
• The approach demonstrates flexibility with heterogeneous base classifiers and addresses data scarcity.

Conclusions:

• The ITO algorithm successfully produces HAHR binary classifiers by integrating diverse base models.
• The proposed method offers a robust solution for scenarios demanding both high accuracy and reliability.
• Results are comparable to established benchmarks like MAQC-II, highlighting the algorithm's effectiveness.