Improving the quality of predictive models in small data GSDOT: A new algorithm for generating synthetic data.

Area of Science:

• Machine Learning
• Data Science
• Artificial Intelligence

Background:

• Many critical domains are limited to small datasets, hindering supervised learning performance.
• Insufficient training data is a major challenge for machine learning algorithm accuracy.
• Existing artificial data generation methods often fall short for small datasets.

Purpose of the Study:

• To address the challenge of small datasets in supervised learning.
• To propose a novel algorithm for generating high-quality artificial data instances.
• To improve the performance of machine learning models trained on limited data.

Main Methods:

• Developed the Geometric Small Data Oversampling Technique (GSDOT).
• GSDOT generates artificial instances within geometric regions around existing data points.
• The generated instances augment the training dataset for supervised learning.

Main Results:

• Significant improvements in model accuracy were observed using GSDOT.
• The proposed technique outperformed models trained on the original small dataset.
• GSDOT showed superior performance compared to other popular artificial data generation methods.

Conclusions:

• The Geometric Small Data Oversampling Technique effectively mitigates the small data problem.
• GSDOT offers a viable solution for enhancing supervised learning in data-scarce environments.
• This approach has the potential for broad impact across various data-limited application domains.