The Benchmarking Native Multi-Output, Regressor Chain and TPOT-MTR Models

Native multi-output expands its capabilities to include multi-target regression using a genetic programming approach. It outperforms state-of-the-art methods on multi-target regression datasets, enhancing prediction accuracy and decision-making in multi-target regression domains. However, it relies heavily on a single target strategy, which may not capture complex interdependencies between multiple targets. Additionally, the systems lack understanding of how linked targets in multi-target regression are handled, making it difficult for practitioners to determine the real connections. The purpose of this study is to evaluate and assess the Native Multi-Output, Regressor Chain and TPOT-MTR model using six relevant public datasets. The methods used were Pearson Correlation and aRRMSE. The research focuses on multi-target regression using AutoML and TPOT, focusing on the TPOT multi-output regression technique. The proposed model is expected to improve correlation and provide customization options, contributing to the advancement of multi-target regression using AutoML techniques. It concludes with a comprehensive analysis of the algorithm’s performance in addressing difficulties in multiple target regression and evaluating its ability to identify and employ relevant features for improved predictive correlation. The results show that TPOT-MTR shows stronger correlation performance across most datasets, especially when target relationship capture is vital. However, its performance advantage is reduced in datasets with weakly connected or naturally well-structured targets. In conclusion, TPOT-MTR is a reliable tool for modelling target correlations, especially in multi-target learning tasks. Its performance varies depending on the dataset structure, with Regressor Chain and Native Multi-Output performing better in densely linked datasets. However, it may not always provide the lowest aRRMSE, making it a good substitute for simpler models. Further research can explore more intricate datasets and integrate neural networks.