A Perspective On Test Methodologies For Time Series Forecasting As Supervised Learning

Abstract

Supervised Machine Learning algorithms are widely used to predict the outcomes of the future dataset. It is the learning method to map from input values to output values. The aim is to predict the output values approximately when new input datasets are given. We discuss test methods on time series forecasting as supervised learning in this work. In time series forecasting, a sequence of time series datasets is rearranged to form like supervised learning problem. This model is trained using input as previous timestamp and output as next timestamp. Therefore, testing time series forecasting is demanded to find improper rearrangement of dataset and window width in sliding window method to ensure accurate prediction. Sliding window method uses the previous timestamp to predict the next time step. The added previous time steps by increase window width are required to check relevant dataset to reduce the false prediction.