The present invention provides a method and system for ensemble based time-series forecasting. The method comprises, receiving a time-series data set as an input, creating a copy of the time-series data set, performing first order differencing on one of the time-series data sets, removing outliers from the time-series data sets, partitioning the time-series data sets into training data sets and validation data sets, training one or more forecast models using training data sets, generating forecast data sets against validation data sets, reverse differencing one of the forecast data sets, removing biasness from the forecast data sets, computing multiple accurac y metrics using validation data sets, updating the forecast models, generating forecast values for a specified forecast horizon, reverse differencing one set of forecast values, removing biasness in the forecast values, obtaining final forecast data set by using the normalized weights of the forecast models, which are computed using multiple accuracy metrics.
展开▼