[2Q1-J-2] Machine learning: models for prediction

[2Q1-J-2-04] Comparison of Deep Time Series Prediction Models Utilizing Linear Regression Model

〇Jun Hozumi¹, Yutaka Matsuo¹ (1. the University of Tokyo)

Keywords:Deep Learning, Time Series

Time series prediction is important for the industry and various prediction models have been proposed. Recently some reports showed that deep learning models have higher accuracy than traditional models. However, it is known that these models take longer time to learn and are difficult to maintain long-term periodicity. To deal with these problems, some researches on deep models with the concepts of autoregressive models have been proposed. However, since these models have not been compared with the same data and same settings, it is unclear which models are effective for diffrent tasks and different data. In this research, we apply these models to different types of time series data under different settings and examine the learning processes and results to capture the features of each model. The results confirmed the claimed merit of the model and suggested that simpler deep models are effective under complicated problems.