TPOT-NN: augmenting tree-based automated machine learning with neural network estimators.

Area of Science:

• Artificial Intelligence
• Machine Learning
• Computer Science

Background:

• Automated machine learning (AutoML) and artificial neural networks (ANNs) are powerful AI tools.
• Limited guidance exists on choosing between AutoML and ANNs, and few tools integrate both.
• Combining AutoML and ANNs can leverage their respective strengths for complex tasks.

Purpose of the Study:

• To introduce TPOT-NN, an extension for the TPOT AutoML software.
• To explore the performance of AutoML augmented with neural network estimators (AutoML+NN).
• To compare AutoML+NN with standard non-NN AutoML for binary classification tasks.

Main Methods:

• Developed TPOT-NN, an extension for the tree-based AutoML software TPOT.
• Evaluated TPOT-NN on public benchmark datasets for binary classification.
• Compared the classification accuracy of AutoML+NN against standard tree-based AutoML.

Main Results:

• TPOT-NN demonstrated effectiveness in improving classification accuracy on certain datasets.
• AutoML+NN achieved comparable or superior accuracy to standard AutoML.
• No loss in accuracy was observed on datasets where standard AutoML performed well.

Conclusions:

• TPOT-NN is a valuable tool for enhancing AutoML capabilities with neural networks.
• Preliminary guidelines for AutoML+NN analyses are provided.
• Future research directions for AutoML+NN methods, particularly within TPOT, are recommended.