Model Selection in Feedforward Neural Networks for Forecasting Inflow and Outflow in Indonesia

Suhartono, Suhartono; Saputri, Prilyandari Dina; Amalia, Farah Fajrina; Prastyo, Dedy Dwi; Ulama, Brodjol Sutijo Suprih

doi:10.1007/978-981-10-7242-0_8

Cited by 12 publications

(5 citation statements)

References 13 publications

Supporting

Mentioning

Contrasting

Unclassified

Order By: Relevance

“…where 1 g , 2 g are the activation functions, w are the estimated parameters that connect the input to the hidden layer, v are the estimated parameters that connect the hidden to the output [22].…”

Section: Neural Networkmentioning

confidence: 99%

Preventing recession through GDP growth prediction: A classical and machine learning classification approach

Saputri,

Angrenani,

Fitriana

2023

Data Science: J. of Computing and Appl. Informatics

View full text Add to dashboard Cite

Classification methods are a popular method applied in many various fields of science. To represent the effect of predictor factors on categorical response variables, different machine learning classification algorithms are used, namely logistic regression, neural network (NN), random forest, support vector machine (SVM), and bayesian model averaging (BMA). Every classifier has its unique characteristic, performing well in certain datasets but not in others. Hence, it is always a quest to find the best classifier to use for a certain dataset. Economic growth, most commonly using a gross regional domestic product, is experiencing a recession or acceleration, especially before and during the COVID-19 pandemic. This research proposed a comparison of classification methods using regional GDP data for 2019-2020, before and during the COVID-19 pandemic, by predictor variables; percentage of workers, foreign direct investment (PMA), regional revenue (PAD), general allocation fund (DAU), revenue sharing fund (DBH), and the dummy of COVID-19. The results are that all selected machine learning models can classify the regional GDP growth perfectly for the training data, but, NN model outperforms the other methods with an accuracy of 100% in training and testing data. COVID-19 and the PMA are the most significant variables predicting regional GDP growth for all models. Further research relating to interpretable machine learning, such as feature interaction, global surrogate, and Shapley values, is also necessary to predict regional GDP growth using machine learning methods.

show abstract

Section: Neural Networkmentioning

confidence: 99%

Preventing recession through GDP growth prediction: A classical and machine learning classification approach

Saputri,

Angrenani,

Fitriana

2023

Data Science: J. of Computing and Appl. Informatics

View full text Add to dashboard Cite

show abstract

“…Dalam penelitian ini, model yang digunakan untuk proyeksi (forecast) total pembayaran non-tunai, yaitu ARIMA, ARIMAX, dan hybrid ARIMAX-ANN akan dibandingkan akurasinya. Untuk memperoleh ramalan ANN, dua pendekatan untuk menyeleksi jumlah variabel input turut dibandingkan, yaitu berdasarkan signifikansi lag residual Autocorrelation Function (ACF) dan regresi stepwise [14]. Di Indonesia, belum ditemukan penelitian serupa yang menggunakan stepwise untuk menyeleksi input.…”

Section: Pendahuluanunclassified

“…Artificial Neural Network (ANN) merupakan metode yang dikembangkan dengan konsep mereplika jaringan syaraf biologis ke dalam model matematis [14]. Salah satu model ANN yang digunakan untuk peramalan adalah feed-forward neural network (FFNN).…”

Section: Arimax-annunclassified

See 1 more Smart Citation

Peramalan Transaksi Pembayaran Non-Tunai Menggunakan Arimax-Ann Dengan Konfigurasi Kalender

Putera

2020

BAREKENG

View full text Add to dashboard Cite

Akses internet yang luas mendorong kian seringnya sistem pembayaran non-tunai digunakan. Di Indonesia, berbagai aktivitas dan transaksi ekonomi seringkali dipengaruhi oleh pergerakan kalender, terutama kalender Hijriyah. Tujuan penelitian ini untuk memodelkan dan meramalkan total pembayaran non-tunai di Indonesia dengan menambahkan konfigurasi kalender sebagai variabel. Digunakan metode ARIMA, ARIMAX dan hibrida ARIMAX-ANN yang akan dibandingkan akurasinya. Diperoleh model terbaik untuk peramalan jumlah pembayaran non-tunai adalah ARIMAX-ANN dengan RMSE terkecil, yaitu Rp 20,9 triliun. Spesifikasi model terbaik tersebut adalah ARIMAX(2,1,1) yang dihibrida dengan ANN yang inputnya diseleksi melalui regresi stepwise. Selain memenuhi asumsi galat yang identik, independen, dan berdistribusi normal, ARIMAX-ANN juga mampu mengikuti dinamika dan tren dari pembayaran non-tunai, khususnya pada bulan jatuhnya Idul Fitri dan periode akhir tahun.

show abstract

“…After obtaining data from each scenario in the first stage, the next step is to model the data from each scenario using ARIMAX and hybrid QRNN model. The study about procedure to choose input in Neural Network with application to currency inflow and outflow forecasting in Indonesia was done by Suhartono et al [19]. The RMSE and MdAE criteria are used to determine the best model for predicting data in each scenario.…”

Section: Simulation Studymentioning

confidence: 99%

Comparison Between Hybrid Quantile Regression Neural Network and Autoregressive Integrated Moving Average With Exogenous Variable for Forecasting of Currency Inflow and Outflow in Bank Indonesia

Prastyo

Suhartono

Puka³

et al. 2018

Jurnal Teknologi

Self Cite

View full text Add to dashboard Cite

Some problems arise in time series analysis are nonlinearity and heteroscedasticity. Methods that can be used to analyze such problems are neural network and quantile regression. There are a lot of studies and developments on both methods, but the study that focuses on the performances of combination of these two methods applied in real case are still limited. Therefore, this study performed a comparison between hybrid Quantile Regression Neural Network (QRNN) and Autoregressive Integrated Moving Average with Exogenous Variable (ARIMAX). Both methods were employed to model the currency inflow and outflow from Bank Indonesia in Nusa Tenggara Timur province. Based on the empirical result, the hybrid QRNN method provided better forecasting for currency outflow whereas the ARIMAX resulted in better forecasting for the inflow.

show abstract

Model Selection in Feedforward Neural Networks for Forecasting Inflow and Outflow in Indonesia

Cited by 12 publications

References 13 publications

Preventing recession through GDP growth prediction: A classical and machine learning classification approach

Preventing recession through GDP growth prediction: A classical and machine learning classification approach

Peramalan Transaksi Pembayaran Non-Tunai Menggunakan Arimax-Ann Dengan Konfigurasi Kalender

Comparison Between Hybrid Quantile Regression Neural Network and Autoregressive Integrated Moving Average With Exogenous Variable for Forecasting of Currency Inflow and Outflow in Bank Indonesia

Contact Info

Product

Resources

About