E-commerce Time Series Forecasting using LSTM Neural Network and Support Vector Regression
| Autor: | Ghassen Chniti, Houda Bakir, Hédi Zaher |
|---|---|
| Rok vydání: | 2017 |
| Předmět: |
0209 industrial biotechnology
Multivariate statistics Artificial neural network Mean squared error business.industry Computer science Deep learning Univariate 02 engineering and technology Machine learning computer.software_genre Support vector machine 020901 industrial engineering & automation Recurrent neural network 0202 electrical engineering electronic engineering information engineering 020201 artificial intelligence & image processing Artificial intelligence Time series business computer |
| Zdroj: | BDIOT |
| DOI: | 10.1145/3175684.3175695 |
| Popis: | The purpose of this paper is to provide a robust forecasting model to predict phone prices on European markets using Long Short Term Memory neural network (LSTM) and Support Vector Regression (SVR). We propose a comparison study of time series forecasting models for these two techniques. The Long Short Term Memory, due to its architecture, is considered as a perfect solution to problems not solvable by classic Recurrent Neural Networks (RNN). On the other hand, Support Vector machines is a very powerful machine learning method for both classification and regression. After studying and comparing several univariate models, SVR and LSTM neural network appear to be the most accurate ones. In addition, we compared multivariate models for both techniques SVR and LSTM. Considering the multivariate approach, by introducing more variables, we obtain better performance for the prediction. In fact, SVR model is able to predict the next day price with RMSE 35.43 Euros with univariate model. However, using multivariate models, recurrent neural network LSTM gives the most accurate prediction for the next day's price with RMSE of 24.718 Euros. |
| Databáze: | OpenAIRE |
| Externí odkaz: |
|