Construction and Factor Analysis of Decision Tree-Based Prediction Models for Automatic Number Estimation of Working Vehicles in Machinery Unloading and Installation

Abstract

This study validates models for predicting the number of working vehicles required for machinery unloading and installation. Constructing decision trees using a dataset of 68 samples from Japan and Thailand resulted in achieving macro-averaged F1 scores of 0.8 or higher on average across all models for the training, validation, and test data. Additionally, important features aligned with expert usage conditions were confirmed. However, the prediction performance for the 10-24t industry transport trolley was lower than that of the others and inconsistent with expert usage conditions. We plan to develop models that can capture complex relations in future studies.