Web application for time-series analysis based on particle filter available on cloud computing system

We develop web application “CloCK-TiME” (Cloud Computing Kernel for Time-series Modeling Engine), which enables users to analyze their time-series data by using a networked PC cluster in a cloud computing system. This software decomposes a given multivariate time-series data into trend, seasonal, autoregressive (AR), and observation noise components, by using the particle filter (PF) algorithm. We also develop a user interface, by which users can set parameters needed in the analysis such as trend order, seasonal period, AR order, and the number of particles. We show an application example in the case of tide gauge data recorded along the coastline of Japan. We are planning to improve our analysis engine in order to obtain not only optimum model parameters but also their posterior distributions eventually by a hybrid method consisting of the PF and the MCMC algorithms.