[3Xin2] Poster session 1

[3Xin2-03] Expanding the Applicability of Decision Tree Learning Using an Annealing Machine

〇Taiki Morinaga¹, Koichiro Yawata¹, Akinori Asahara¹ (1.Hitachi, Ltd.)

Keywords:decision tree, optimization, annealing, quantum, chemistry

The decision tree, a machine learning algorithm, predicts the target variable by classifying data based on conditions for explanatory variables and learns by minimizing prediction error. However, performing an exhaustive search for all possible conditions becomes exponentially time-consuming relative to the number of explanatory variables. Thus, learning usually involves sequentially searching for conditions for each variable, which may lead to locally optimal solutions. To address this, the Quadratic Unconstrained Binary Optimization (QUBO) decision tree has been proposed. This method reduces the problem to a QUBO problem and solves it using an annealing machine, which is capable of rapidly solving QUBO problems. While the QUBO decision tree has achieved higher prediction accuracy than existing decision trees, it is limited to handling binary explanatory variables. This study extends the QUBO decision tree to accommodate real-valued explanatory variables. By applying this method to open data of organic compounds with real-valued explanatory variables, we achieved superior accuracy compared to traditional decision trees.