Meta-learning for time-series forecasting

Abstract

Time series forecasting has been applied in several real word problems to support the decision-making process, for example Electroencephalogram (EEG) analysis, energy consumption, financial stock market and others. However these experiments usually demand long time to select the best model for decision making. Also for this domain need specific knowledge of the time-serie kind for making accurate forecast. The meta-learning objective is select promising algorithms for new datasets based on meta-dataset knowledge discovery. This project aims to develop a meta-learning based recommendation system of statistical and machine learning models for time-series forecasting. Allied to previous will make an experimental analysis of existing meta-features in time-series and regression domain. (AU)