Simulation Study for Determining the Best Architecture of Multilayer Perceptron for Forecasting Nonlinear Seasonal Time Series

Suhartono, Suhartono; Amalia, Farah Fajrina; Saputri, Prilyandari Dina; Rahayu, Santi Puteri; Ulama, Brodjol Sutijo Suprih

doi:10.1088/1742-6596/1028/1/012214

Cited by 4 publications

(2 citation statements)

References 14 publications

Supporting

Mentioning

Contrasting

Order By: Relevance

“…Zhang et al (2001) specifically simulate non-linear time series using non-linear DGPs such as Sign AR processes, TAR models, NAR models, Non-linear Moving Average (NMA) models and STAR models to assess the effect from the input nodes, hidden nodes of NNs, and sample size on the forecasting accuracy of FFNNs in comparison with ARIMA models. Similar to this, Suhartono et al (2018) also perform a study using non-linear seasonal time series simulated with an Exponential Smooth Transition Auto-Regressive (ESTAR) model to explore determining factors such as the inputs and the number of neurons in the hidden layers on the FFNN's forecasting accuracy.…”

Section: Data Generating Processesmentioning

confidence: 99%

Global Models for Time Series Forecasting: A Simulation Study

Hewamalage¹,

Bergmeir²,

Bandara³

2020

Preprint

View full text Add to dashboard Cite

In the current context of Big Data, the nature of many forecasting problems has changed from predicting isolated time series to predicting many time series from similar sources. The availability of large sets of time series has opened up the opportunity to develop competitive global forecasting models that simultaneously learn from many time series, which have shown recently promising results in forecasting competitions, outperforming state-of-the-art univariate forecasting techniques. Nevertheless, it still remains unclear under which circumstances global forecasting models can outperform the univariate benchmarks, especially along the dimensions of the homogeneity/heterogeneity of series, the complexity of patterns in the series, the complexity of forecasting models, and the lengths/number of series. Our study attempts to address this problem through investigating the effect from these factors, by simulating a number of datasets that have controllable time series characteristics. Specifically, we simulate time series from simple data generating processes (DGP), such as Auto Regressive (AR) and Seasonal AR, to complex DGPs, such as Chaotic Logistic Map, Self-Exciting Threshold Auto-Regressive, and Mackey-Glass Equations. The data heterogeneity is introduced by mixing time series generated from several DGPs into a single dataset. The lengths and the number of series in the dataset are varied in different scenarios. Unlike in real-world datasets, the simulated environments provide complete control over the underlying datasets. We perform experiments on these datasets using global forecasting models including Recurrent Neural Networks (RNN), Feed-Forward Neural Networks, Pooled Regression (PR) models and Light Gradient Boosting Models (LGBM), and compare their performance against standard statistical univariate forecasting techniques. Our experiments demonstrate that when trained as global forecasting models, techniques such as RNNs and LGBMs, which have complex non-linear modelling capabilities, are competitive methods in general under challenging forecasting scenarios such as series having short lengths, datasets with heterogeneous series and having minimal prior knowledge of the patterns of the series. This makes these techniques promising candidates for forecasting under uncertain situations as opposed to techniques such as PR and AR models, which assume linearity of the underlying data.

show abstract

Section: Data Generating Processesmentioning

confidence: 99%

Global Models for Time Series Forecasting: A Simulation Study

Hewamalage¹,

Bergmeir²,

Bandara³

2020

Preprint

View full text Add to dashboard Cite

show abstract

“…Initially, some classical time series methods such as time series regression, ARIMA and ARIMAX were applied for inflow and outflow forecasting (Setiawan et al 2015). Then, several researchers used nonlinear and machine learning methods such as Feed-forward Neural Network or FFNN, and hybrid Quantile Regression Neural Network or QRNN (Suhartono et al 2018a). Recently, a hybrid Singular Spectrum Analysis and Neural Network or SSA-NN also be employed for inflow and outflow forecasting in Indonesia (Suhartono et al 2017).…”

Section: Introductionmentioning

confidence: 99%

Deep Neural Network for Forecasting Inflow and Outflow in Indonesia

Suhartono¹,

Ashari²,

Prastyo³

et al. 2019

JSM

View full text Add to dashboard Cite

An optimal planning in the preparation of Money Requirement Plan (MRP) by Bank Indonesia is highly beneficial to maintain the availability of money in the community. One of the main factors needed in preparing of MRP is an accurate information about inflow and outflow. This study is to apply Deep Neural Network (DNN) for forecasting inflow and outflow in Indonesia and to compare its performance to ARIMAX as a simpler method and hybrid Singular Spectrum Analysis and DNN (SSA-DNN) as a more complex method. This study focuses on determining the best inputs in DNN, particularly for forecasting time series. A simulation study is used for evaluating the performance of each method related to the patterns in the time series. The real data are monthly inflow and outflow on 5 banknotes denominations from January 2003 to December 2016. The performance was evaluated based on Root Mean Square Error Prediction and Symmetry Mean Absolute Percentage Error Prediction criteria. The results of the simulation study showed that DNN yielded a more accurate forecast than ARIMAX and hybrid SSA-DNN in predicting time series with a trend, seasonal, calendar variation, and nonlinear noise patterns. Moreover, the results of inflow and outflow forecasting showed that DNN provided a more accurate prediction on most all banknotes denominations compared to ARIMAX and hybrid SSA-DNN. In general, these results show that DNN as machine learning model outperforms both ARIMAX as a simpler statistical model and hybrid SSA-DNN as a more complex model.

show abstract

Neural Network Models for Time Series Analysis and Estimation

Nuaimy¹

2023

Lecture Notes in Networks and Systems

View full text Add to dashboard Cite

Simulation Study for Determining the Best Architecture of Multilayer Perceptron for Forecasting Nonlinear Seasonal Time Series

Cited by 4 publications

References 14 publications

Global Models for Time Series Forecasting: A Simulation Study

Global Models for Time Series Forecasting: A Simulation Study

Deep Neural Network for Forecasting Inflow and Outflow in Indonesia

Neural Network Models for Time Series Analysis and Estimation

Contact Info

Product

Resources

About