Improving modeling of stochastic processes by smart denoising

Abstract

This paper proposes a novel method for modeling stochastic processes, which are known to be notoriously hard to predict accurately. State of the art methods quickly overfit and create big differences between train and test datasets. We present a method based on smart noise addition to the data obtained from unknown stochastic process, which is capable of reducing data overfitting. The proposed method works as an addition to the current state of the art methods in both supervised and unsupervised setting. We evaluate the method on equities and cryptocurrency datasets, specifically chosen for their chaotic and unpredictable nature. We show that with our method we significantly reduce overfitting and increase performance, compared to several commonly used machine learning algorithms: Random forest, General linear model and LSTM deep learning model.