A mapping study of ensemble classification methods in lung cancer decision support systems

Area of Science:

• Medical Informatics
• Artificial Intelligence
• Computational Biology

Background:

• Accurate medical data classification is crucial for developing effective clinical decision support systems.
• Ensemble classification methods offer improved performance over single classifiers in artificial intelligence applications.
• Lung cancer detection research requires robust and accurate classification techniques.

Purpose of the Study:

• To systematically map the state-of-the-art ensemble classification methods used in lung cancer detection.
• To identify trends, common practices, and research gaps in this domain.
• To provide insights for future research directions in lung cancer decision support systems.

Main Methods:

• A systematic mapping study was conducted, reviewing 65 papers published between 2000 and 2018.
• Automatic searches were performed across four major digital libraries, followed by a rigorous selection process.
• Analysis focused on classification tasks, ensemble construction techniques, and combination rules.

Main Results:

• Diagnosis emerged as the most frequently studied task in lung cancer detection using ensemble methods.
• Homogeneous ensembles and decision trees were the most prevalent approaches for constructing ensembles.
• The majority voting rule was the predominant method for combining classifier outputs.
• Parameter tuning of ensemble techniques was infrequently addressed in the reviewed literature.

Conclusions:

• The study highlights the dominance of diagnosis as a task and specific methods like homogeneous ensembles and majority voting in lung cancer detection.
• Identified gaps include limited exploration of parameter tuning, heterogeneous ensembles, and novel combination rules.
• Future research should focus on addressing these gaps to advance lung cancer classification and decision support systems.