An Approach to Growth Delimitation of Straight Line Segment Classifiers Based on a Minimum Bounding Box.

Area of Science:

• Computer Science
• Machine Learning
• Pattern Recognition

Background:

• Supervised machine learning algorithms are widely used for binary classification.
• The straight-line segment classifier offers low complexity and competitive performance.
• Unconstrained segment lengths during training can negatively impact classification accuracy.

Purpose of the Study:

• To propose a novel approach for adjusting straight-line segment classifier parameters.
• To constrain the search space during training to restrict segment lengths.
• To enhance the reliability and classification rates of the straight-line segment classifier.

Main Methods:

• Developed a constrained optimization approach for segment extremity placement.
• Introduced a scale factor parameter to tune the search space and limit segment lengths.
• Evaluated the approach on ten artificial and eight UCI Machine Learning Repository datasets.

Main Results:

• The proposed constrained approach demonstrated promising results compared to other classifiers.
• Limiting segment lengths positively impacted the classification rate.
• The adjusted classifier maintained reliable performance across diverse datasets.

Conclusions:

• The constrained straight-line segment classifier is a reliable tool for binary classification.
• Its straightforward interpretation and strong classification rates make it suitable for industrial decision-making.
• The approach effectively addresses the issue of excessive segment lengths in classifier training.