APPLICATION AND PERFORMANCE COMPARISON OF MULTI-OUTPUT MACHINE LEARNING FOR NUMERICAL-NUMERICAL AND NUMERICAL-CATEGORICAL OUTPUTS

Multi-Output Machine Learning is an advancement of traditional machine learning, designed to predict multiple output variables simultaneously while considering the relationships between these output variables. Multi-Output Machine Learning is essential as a decision support tool because decision-making in many problems generally considers multiple factors. The use of Multi-Output Machine Learning is more advantageous than conventional machine learning in terms of time efficiency, addressing data limitations, and ease of maintenance. These benefits will significantly impact cost savings for industries utilizing Big Data. The models used in this research include Multivariate Regression Tree, Multivariate Random Forest, and Multi-Output Neural Network. The Multivariate Regression Tree and Multivariate Random Forest are developed by modifying the splitting function using Mahalanobis distance. The topological changes introducing shared and private hidden layers are the key development of the Multi-Output Neural Network. The prediction results indicated a trade-off in error between two output variables when comparing the Multivariate Regression Tree and Multivariate Random Forest with their single output counterparts. Meanwhile, the Multi-Output Neural Network model successfully improved the prediction results for both output variables. This research also introduces Mixed Multi-Output Machine Learning, which can predict numerical and categorical output variables. The Mixed Multi-Output Machine Learning model utilizes the logit values from the Logistic Regression model to extend the range of prediction results beyond the 0 to 1 interval. Multi-Output Neural Network is the sole model that produces predictions with relatively small errors and high accuracy values.