Multi-Objective Evolutionary Instance Selection for Regression Tasks

Area of Science:

• Machine Learning
• Data Mining

Background:

• Instance selection is crucial for reducing data size and improving model efficiency.
• Erroneous and redundant data can negatively impact regression model performance.

Purpose of the Study:

• To develop an effective instance selection method for regression tasks.
• To optimize the trade-off between dataset compression and prediction accuracy.

Main Methods:

• Application of the NSGA-II multi-objective evolutionary algorithm for subset selection.
• Utilizing the k-NN algorithm to evaluate the quality of selected subsets.
• Analysis of various parameters influencing the instance selection process.

Main Results:

• The proposed method generates a Pareto front of optimal solutions, balancing RMSE and compression.
• Demonstrated effectiveness in minimizing both prediction error (RMSE) and dataset size.
• Efforts made to reduce the computational complexity of the approach.

Conclusions:

• The developed instance selection technique is efficient for regression tasks.
• Achieves a favorable balance between data reduction and predictive performance.
• Offers a valuable tool for optimizing large datasets in machine learning.