Abstract
This study investigates the effectiveness of probability forecasts output by standard machine learning techniques (Neural Network, C4.5, K-Nearest Neighbours, Naive Bayes, SVM and HMM) when tested on time series datasets from various problem domains. Raw data was converted into a pattern classification problem using a sliding window approach, and the respective target prediction was set as some discretised future value in the time series sequence. Experiments were conducted in the online learning setting to model the way in which time series data is presented. The performance of each learner's probability forecasts was assessed using ROC curves, square loss, classification accuracy and Empirical Reliability Curves (ERC) [1]. Our results demonstrate that effective probability forecasts can be generated on time series data and we discuss the practical implications of this.
Sign-in/Register to access full text options 
Talk to us
Join us for a 30 min session where you can share your feedback and ask us any queries you have
Schedule a callDisclaimer: All third-party content on this website/platform is and will remain the property of their respective owners and is provided on "as is" basis without any warranties, express or implied. Use of third-party content does not indicate any affiliation, sponsorship with or endorsement by them. Any references to third-party content is to identify the corresponding services and shall be considered fair use under The CopyrightLaw.
