The objective of this project is to develop models based on soft computing and data mining for time series forecasting. There are many models such as support vector machines or association rules, providing good results in a great number of world real problems. Association rules have been traditionally used to obtain static relationships between different variables. However, these models have not been exploited when time variables playing a crucial role. Therefore, the use of association rules for time series forecasting lead to a new paradigm in the prediction field. Thus, a prediction based on exact values at a future timestamp will not be obtained as in the classic approach, but a band of values within which it is predicted that the time series will remain in the future will be obtained. On the other hand, vector support machines have been mainly used in static classification problems, being the time series classification a wide study field with many applications and of great interest nowadays.
Finally, the possibility of extending models developed for time series forecasting to the outlier prediction, such as earthquakes or certain high levels of air pollution, is proposed as a second objective of this project. It is important to model phenomena of nature but also to model atypical behaviors with the aim of predicting such anomalies before they happen. A great number of applications already exist in this context and will grow considerably in the future.
In this project, the results will focus on two types of temporal data: earthquake data and environmental data. Currently, due to the recent earthquakes occurred in Spain or in Chile, obtaining a forecast of when and where these natural disasters may occur will be a big step forward with a high social and economic impact. On the other hand, the prediction of atmospheric pollutants is a hot topic in the worldwide and can be considered strategic for the sustainable development of the Andalusian region.