FMarkNet: Forecasting Model Based on Neural Networks and the Markowitz Model

The forecasting problem is vital in many areas, such as Energy, industry, financial series, and climate change. In forecasting, an ensemble is a combination of methods with different approaches. In other words, selecting only neural networks in an ensemble is not advisable because the advantages of various strategies are not exploited. This chapter presents FMarkNet, a forecasting neural model based on the famous Market Markowitz Portfolio Model. To this end, it is essential to select the best individual methods for the application area and their metrics error for each time series. For this work, we used M4 competition methods and the typical error metrics applied to it. We tested the FMarkNet with a dataset of temperature for climate change and achieved an excellent performance. The MAPE results using a CNN alone was 4.98%, and when we included additional algorithms, the ensemble achieved a MAPE of 3.01%.