A just-in-time adaptive classification system based on the intersection of confidence intervals rule

Area of Science:

• Machine Learning
• Data Science
• Statistical Modeling

Background:

• Classification systems struggle in nonstationary environments where data generation processes change.
• Existing instance selection methods involve replacing obsolete data with new samples for retraining.

Purpose of the Study:

• To propose an adaptive classifier for nonstationary environments.
• To detect changes in data generation processes and identify new data for classifier configuration.

Main Methods:

• Developed an adaptive classifier using the intersection of confidence intervals rule.
• Implemented a method to detect process changes and select novel data samples.
• No assumptions were made about the underlying data generation distribution.

Main Results:

• The proposed adaptive classification system demonstrated effectiveness in nonstationary environments.
• The system successfully detected abrupt changes in the data generation process.
• New data was accurately identified for classifier retraining.

Conclusions:

• The adaptive classifier is particularly effective for processes with abrupt changes.
• The method offers a robust approach to classification in dynamic environments.
• This research contributes to the development of self-adapting machine learning systems.