A Novel Classification Method: Neighborhood-Based Positive Unlabeled Learning Using Decision Tree (NPULUD).

Area of Science:

• Machine Learning
• Data Science
• Artificial Intelligence

Background:

• Standard binary classification requires both positive and negative samples, which is not always feasible.
• Certain applications lack sufficient labeled data, necessitating alternative learning strategies.

Purpose of the Study:

• To introduce a novel method for positive and unlabeled (PU) learning called neighborhood-based positive unlabeled learning using decision tree (NPULUD).
• To address the challenge of building classification models when only positive samples are available.

Main Methods:

• NPULUD utilizes a nearest neighborhood approach for PU strategy.
• A decision tree algorithm is employed for classification, leveraging entropy measures to assess data uncertainty.

Main Results:

• The NPULUD method achieved an average accuracy of 87.24% across 24 real-world datasets.
• This represents a statistically significant improvement of 7.74% compared to state-of-the-art methods.
• The proposed method outperformed traditional supervised learning, which achieved an average accuracy of 83.99%.

Conclusions:

• NPULUD offers an effective solution for classification tasks with limited labeled data.
• The integration of neighborhood-based strategies and decision trees with entropy significantly enhances PU learning performance.
• The method demonstrates robust performance and statistical superiority over existing techniques.