There have been many studies seeking to predict excess returns in financial time series data. Nevertheless, not many studies have focused on applying machine learning approaches among factors in different asset classes. The main objective of this paper is to analyze whether a predictability of return in financial products could be improved by considering factors obtained from other asset classes, and to indicate the effectiveness of machine learning in financial time series prediction. We targeted 10-year Japanese Government Bond(10-year JGB), and Nikkei Stock Average Index, implementing non-linear machine learning approaches as well as conventional multiple linear regression models to predict returns in both assets. The results suggest that considering factors from other asset classes could improve return prediction both in 10-year JGB and Nikkei Stock Average, especially when using non-linear approaches.