We present a neural network technique for the analysis and extrapolation of time-series data called Neural Decomposition (ND). Units with a sinusoidal activation function are used to perform a Fourier-like decomposition of training samples into a sum of sinusoids, augmented by units with nonperiodic activation functions to capture linear trends and other nonperiodic components. We show how careful weight initialization can be combined with regularization to form a simple model that generalizes well. Our method generalizes effectively on the Mackey-Glass series, a dataset of unemployment rates as reported by the U.S. Department of Labor Statistics, a time-series of monthly international airline passengers, the monthly ozone concentration in downtown Los Angeles, and an unevenly sampled time-series of oxygen isotope measurements from a cave in north India. We find that ND outperforms popular time-series forecasting techniques including ARIMA, SARIMA, SVR with a radial basis function, Gashler and Ashmore’s model, and echo state networks.
","abstract_html":"<p>We present a neural network technique for the analysis and extrapolation of time-series data called Neural Decomposition (ND). Units with a sinusoidal activation function are used to perform a Fourier-like decomposition of training samples into a sum of sinusoids, augmented by units with nonperiodic activation functions to capture linear trends and other nonperiodic components. We show how careful weight initialization can be combined with regularization to form a simple model that generalizes well. Our method generalizes effectively on the Mackey-Glass series, a dataset of unemployment rates as reported by the U.S. Department of Labor Statistics, a time-series of monthly international airline passengers, the monthly ozone concentration in downtown Los Angeles, and an unevenly sampled time-series of oxygen isotope measurements from a cave in north India. We find that ND outperforms popular time-series forecasting techniques including ARIMA, SARIMA, SVR with a radial basis function, Gashler and Ashmore’s model, and echo state networks.</p>","abstract_has_math":false,"creators":["Godfrey, Luke"],"institution":null,"degree_name":"Master of Science in Computer Science (MS)","degree_level":"Thesis","degree_discipline":null,"degree_department":null,"school":null,"contributors":["Li, Wing Ning","Wu, Xintao"],"advisors":["Gashler, Michael S."],"committee_chairs":[],"committee_members":[],"year":2015,"date_issued":"2015-12-01T08:00:00Z","date_published":"2015-12-01T08:00:00Z","updated_at":"2026-07-24T00:59:01Z","subjects":["Applied sciences","Other Computer Sciences"],"languages":[],"rights":[],"rights_urls":[],"identifier_entries":[]},"links":{"outbound_url":"https://scholarworks.uark.edu/etd/1360","outbound_label":"Repository record","outbound_source":"dc:identifier"},"metadata_groups":[{"id":"people","label":"People","entries":[{"key":"dc:contributor","label":"Contributor","values":["Li, Wing Ning","Wu, Xintao"]},{"key":"dc:contributor.advisor","label":"Advisor","values":["Gashler, Michael S."]},{"key":"dc:creator","label":"Author","values":["Godfrey, Luke"]}]},{"id":"academic_context","label":"Academic Context","entries":[{"key":"dc:date","label":"Dc Date","values":["2015"]},{"key":"dc:date.available","label":"Dc Date Available","values":["2017-09-29T07:00:00Z"]},{"key":"thesis:degree_level","label":"Degree Level","values":["Thesis"]},{"key":"thesis:degree_name","label":"Degree Name","values":["Master of Science in Computer Science (MS)"]}]},{"id":"subjects_keywords","label":"Subjects and Keywords","entries":[{"key":"dc:subject","label":"Dc Subject","values":["Applied sciences","Other Computer Sciences"]}]},{"id":"identifiers","label":"Identifiers","entries":[{"key":"dc:identifier","label":"Identifier","values":["https://scholarworks.uark.edu/etd/1360"]}]},{"id":"additional","label":"Additional Metadata","entries":[{"key":"dc:description.abstract","label":"Abstract","values":["We present a neural network technique for the analysis and extrapolation of time-series data called Neural Decomposition (ND). Units with a sinusoidal activation function are used to perform a Fourier-like decomposition of training samples into a sum of sinusoids, augmented by units with nonperiodic activation functions to capture linear trends and other nonperiodic components. We show how careful weight initialization can be combined with regularization to form a simple model that generalizes well. Our method generalizes effectively on the Mackey-Glass series, a dataset of unemployment rates as reported by the U.S. Department of Labor Statistics, a time-series of monthly international airline passengers, the monthly ozone concentration in downtown Los Angeles, and an unevenly sampled time-series of oxygen isotope measurements from a cave in north India. We find that ND outperforms popular time-series forecasting techniques including ARIMA, SARIMA, SVR with a radial basis function, Gashler and Ashmore’s model, and echo state networks.
"]},{"key":"dc:title","label":"Title","values":["Neural Decomposition of Time-Series Data for Effective Generalization"]}]}],"canonical_facts":{"dc:contributor":["Li, Wing Ning","Wu, Xintao"],"dc:contributor.advisor":["Gashler, Michael S."],"dc:creator":["Godfrey, Luke"],"dc:date":["2015"],"dc:date.available":["2017-09-29T07:00:00Z"],"dc:description.abstract":["We present a neural network technique for the analysis and extrapolation of time-series data called Neural Decomposition (ND). Units with a sinusoidal activation function are used to perform a Fourier-like decomposition of training samples into a sum of sinusoids, augmented by units with nonperiodic activation functions to capture linear trends and other nonperiodic components. We show how careful weight initialization can be combined with regularization to form a simple model that generalizes well. Our method generalizes effectively on the Mackey-Glass series, a dataset of unemployment rates as reported by the U.S. Department of Labor Statistics, a time-series of monthly international airline passengers, the monthly ozone concentration in downtown Los Angeles, and an unevenly sampled time-series of oxygen isotope measurements from a cave in north India. We find that ND outperforms popular time-series forecasting techniques including ARIMA, SARIMA, SVR with a radial basis function, Gashler and Ashmore’s model, and echo state networks.
"],"dc:identifier":["https://scholarworks.uark.edu/etd/1360"],"dc:subject":["Applied sciences","Other Computer Sciences"],"dc:title":["Neural Decomposition of Time-Series Data for Effective Generalization"],"thesis:degree_level":["Thesis"],"thesis:degree_name":["Master of Science in Computer Science (MS)"]},"updated_at":"2026-07-24T00:59:01Z"}