Learning With Incremental Instances and Features.

Area of Science:

• Machine Learning
• Data Mining
• Computer Science

Background:

• Real-world data streams often grow dynamically in volume and feature dimensions.
• Data is frequently collected in batches, creating blocky, trapezoidal structures.
• Existing methods struggle with data streams that have both increasing volume and features.

Purpose of the Study:

• To propose a novel algorithm for learning classification models from blocky trapezoidal data streams.
• To develop dynamic model update strategies for increasing data and expanding feature spaces.
• To address limitations of current algorithms in handling dynamic, block-based data streams.

Main Methods:

• Developed a learning with incremental instances and features (IIF) algorithm.
• Divided data streams into parts and constructed corresponding classifiers.
• Utilized a global loss function for classifier interaction and ensemble methods for the final model.
• Transformed the method into a kernel-based approach for broader applicability.

Main Results:

• The proposed IIF algorithm effectively learns from blocky trapezoidal data streams.
• Dynamic model update strategies successfully handle increasing training data and feature spaces.
• Both theoretical and empirical analyses confirm the algorithm's effectiveness.
• The kernelized version enhances the method's practical applicability.

Conclusions:

• The IIF algorithm provides an effective solution for classification tasks with blocky trapezoidal data streams.
• The method demonstrates robust performance in handling dynamic data growth and feature expansion.
• This work advances the field of stream learning by addressing previously unmet challenges.