Multi-layer Perceptron and Pruning
Journal Title: Turkish Journal of Forecasting - Year 2017, Vol 1, Issue 1
Abstract
A Multi-Layer Perceptron (MLP) defines a family of artificial neural networks often used in TS modeling and forecasting. Because of its “black box” aspect, many researchers refuse to use it. Moreover, the optimization (often based on the exhaustive approach where “all” configurations are tested) and learning phases of this artificial intelligence tool (often based on the Levenberg-Marquardt algorithm; LMA) are weaknesses of this approach (exhaustively and local minima). These two tasks must be repeated depending on the knowledge of each new problem studied, making the process, long, laborious and not systematically robust. In this short communication, a pruning process is presented. This method allows, during the training phase, to carry out an inputs selecting method activating (or not) inter-nodes connections in order to verify if forecasting is improved. We propose to use iteratively the popular damped least-squares method to activate inputs and neurons. A first pass is applied to 10% of the learning sample to determine weights significantly different from 0 and delete other. Then a classical batch process based on LMA is used with the new MLP. The validation is done using 25 measured meteorological TS and crosscomparing the prediction results of the classical LMA and the 2-stage LMA.
Authors and Affiliations
Cyril Voyant, Christophe Paoli, Marie-Laure Nivet, Gilles Notton, Alexis Fouilloy, Fabrice Motte
A New Journal for Forecasting Research
The Turkish Journal of Forecasting (TJF) is an open access international journal and it is published semi-annually. The aim of the TJF is to procure a platform to integrate the research subjects and fields, and to bridge...
An Enhanced Neural-based Bi-Component Hybrid Model for Foreign Exchange Rate Forecasting
Foreign exchange rates are among the most important economic indices in the international monetary markets. For large multinational firms, which conduct substantial currency transfers in the course of business, being abl...
Why and How Does Exponential Smoothing Fail? An In Depth Comparison of ATA-Simple and Simple Exponential Smoothing.
Even though exponential smoothing (ES) is publicized as one of the most successful forecasting methods in the time series literature and it is widely used in practice due to its simplicity, its accuracy can be affected b...
Bayesian Learning based Gaussian Approximation for Artificial Neural Networks
In the nonlinear systems, the pre-knowledge about the exact functional structure between inputs and outputs is mostly either unavailable or insufficient. In this case, the artificial neural networks (ANNs) are useful too...
A Fuzzy Modelling Approach to Robust Design via Loss Functions
Especially in a world where industrial development is reinforced by globalization tendencies, competitive companies know that satisfying customers' needs and running a successful operation requires a process that is reli...