Efficient Cluster-Based Boosting for Semisupervised Classification.

Area of Science:

• Machine Learning
• Data Science
• Artificial Intelligence

Background:

• Semisupervised classification (SSC) leverages both labeled and unlabeled data for improved instance classification.
• Large-scale datasets are common in SSC, posing challenges for existing algorithms.
• Current ensemble methods often struggle with the computational demands of large datasets.

Purpose of the Study:

• To introduce an efficient and scalable semisupervised classification algorithm for large datasets.
• To address the limitations of existing ensemble methods in handling large-scale data.
• To develop a method that achieves high generalization performance with reduced data and computational complexity.

Main Methods:

• Proposed Efficient Cluster-based Boosting (ECB), a multiclass SSC algorithm.
• Incorporated cluster-based regularization to prevent decision boundary generation in dense regions.
• Implemented a semisupervised instance selection procedure to identify informative unlabeled data for base learner training.

Main Results:

• ECB demonstrates good performance using a small subset of selected data and a limited number of base learners.
• Experimental results confirm ECB's scalability to large datasets.
• ECB achieves generalization performance comparable to state-of-the-art methods.

Conclusions:

• ECB offers an efficient and scalable solution for semisupervised classification on large datasets.
• The proposed method effectively handles large-scale data while maintaining high accuracy.
• ECB represents a significant advancement in ensemble-based semisupervised learning.