Table 1 Synopsis of the most relevant...

Table 1

Synopsis of the most relevant works

Paper	Data source	Level	Multi-echelon	Dependent and independent variables	Area of focus	Technique	Verification method	Other important findings
Sun et al. (2018)	Balance of payments statistics, import/export data from 217 countries	Country level, yearly GDP	No	Dependent: GDP Independent: Top 10 import/export countries	Forecasting GDP	LR, RBF, SVM, RD, REP	10-fold cross-validation; MAE RMSE	RBF algorithm outperforms others, linear regression shows good fit
Wohl and Kennedy (2018)	World Bank, COMTRADE, CEPII	Country level, yearly whole import	No	Dependent: Natural log of trade values Independent: Gravity model variables	Forecasting international trade	NN	Out-of-sample predictions; RMSE	Neural network consistently outperforms OLS and PPML, interpretability challenges
Jošić and Žmuk (2022)	Trade Map, CEPII	Country level, yearly whole import	No	Dependent: Bilateral trade flows of Croatia, Independent: Various factors influencing trade	Forecasting Croatia's bilateral trade flows	GP, LR, MLP	Out-of-sample predictions; MAPE	ML algorithms demonstrate good predictive capability
Lei et al. (2022)	Brazil transit and USA air network	State level, monthly	No	Dependent: Edge removal Independent: Population, betweenness	Forecasting edge removal	27 ML algorithms	Out-of-sample validation; Recall, F1 Score	RBF algorithm outperforms other 26 ML algorithms
Mueller (2023)	EU Air network	Country level, monthly	No	Dependent: Edge weight and edge removal Independent: Population, distance	Forecasting edge removal and edge weight	RNN	Out-of-sample validation; AUPRC, RMSE	Weight prediction led to poor predictions
This paper	COMTRADE, IPI and, CIF	Country level, monthly	Yes	Dependent: Flow of materials in different echelons Independent: demand, IPI and, CIF	Forecasting material flows for final products and related materials	Stacked and hybrid ML method	MAE, RMSE, MAPE, R-squared; cross validation and out-of-sample validation	Stacked ML outperforms hybrid ML method

Paper	Data source	Level	Multi-echelon	Dependent and independent variables	Area of focus	Technique	Verification method	Other important findings
Sun et al. (2018)	Balance of payments statistics, import/export data from 217 countries	Country level, yearlyGDP	No	Dependent: GDPIndependent: Top 10 import/export countries	Forecasting GDP	LR, RBF, SVM, RD, REP	10-fold cross-validation; MAE RMSE	RBF algorithm outperforms others, linear regression shows good fit
Wohl and Kennedy (2018)	World Bank, COMTRADE, CEPII	Country level, yearly whole import	No	Dependent: Natural log of trade valuesIndependent: Gravity model variables	Forecasting international trade	NN	Out-of-sample predictions; RMSE	Neural network consistently outperforms OLS and PPML, interpretability challenges
Jošić and Žmuk (2022)	Trade Map, CEPII	Country level, yearly whole import	No	Dependent: Bilateral trade flows of Croatia, Independent: Various factors influencing trade	Forecasting Croatia's bilateral trade flows	GP, LR, MLP	Out-of-sample predictions; MAPE	ML algorithms demonstrate good predictive capability
Lei et al. (2022)	Brazil transit and USA air network	State level, monthly	No	Dependent: Edge removal Independent: Population, betweenness	Forecasting edge removal	27 ML algorithms	Out-of-sample validation; Recall, F1 Score	RBF algorithm outperforms other 26 ML algorithms
Mueller (2023)	EU Air network	Country level, monthly	No	Dependent: Edge weight and edge removalIndependent: Population, distance	Forecasting edge removal and edge weight	RNN	Out-of-sample validation; AUPRC, RMSE	Weight prediction led to poor predictions
This paper	COMTRADE, IPI and, CIF	Country level, monthly	Yes	Dependent: Flow of materials in different echelonsIndependent: demand, IPI and, CIF	Forecasting material flows for final products and related materials	Stacked and hybrid ML method	MAE, RMSE, MAPE, R-squared; cross validation and out-of-sample validation	Stacked ML outperforms hybrid ML method

Note(s): Abbreviations: LR: Linear Regression; RBF: radial basis function; SVM: support vector machine; RD: regression by discretization; REP: reduced error pruning Tree; MAE: mean absolute error; GP: Gaussian Process; NN: neural network; MLP: multilayer perceptron; OLS: ordinary least squares; PPML Poisson pseudo-maximum likelihood; RNN: recurrent neural network and AUPRC: area under the precision-recall curve

[ViewLarge]

Sharing Unavailable