Figure 1. Research framework Source: Authors’ own work A flow diagram illustrates crude oil price prediction using the c G A N-T C N-B I-L S T M framework. The diagram presents crude oil data, green stocks and bonds, blue markets, E S G, and geopolitical risks, which first form a dataset. T... More about this image found in Research framework Source: Authors’ own work A flow diagram illustrat...

Figure 2. Correlation heatmap of the variables (level-based and transformed prices) Source: Authors’ own work Two heatmaps compare correlation matrices for original and transformed variables. The panel 1 heatmap presents correlations for P I O, G N R, I C L N, S P C L E A N, G B, E S G, W T... More about this image found in Correlation heatmap of the variables (level-based and transformed prices) ...

Figure 3. Scatterplot of predicted and actual WTI prices Source: Authors’ own work A scatter plot compares actual W T I and predicted W T I for the hybrid c G A N model. The horizontal axis gives actual W T I values from about 65 to 95, and the vertical axis gives predicted W T I values f... More about this image found in Scatterplot of predicted and actual WTI prices Source: Authors’ own wo...

Figure 4. Prediction error comparison (in levels) Source: Authors’ own work A histogram compares prediction errors across 16 forecasting models. The histogram presents prediction errors in levels, with values ranging from about negative 35 to 16 and frequency values up to about 55. Hybrid c... More about this image found in Prediction error comparison (in levels) Source: Authors’ own work A h...

Figure 5. Predicted WTI prices with different ML models Source: Authors’ own work A line graph compares one-step-ahead W T I forecasts from the hybrid c G A N model and selected models. The horizontal axis represents time from July 2023 to March 2025, and the vertical axis represents W ... More about this image found in Predicted WTI prices with different ML models Source: Authors’ own w...

Figure 6. Prediction error comparison with diverse machine learning models (family 1) Source: Authors’ own work Three panels compare forecasting error distributions, fitted error distributions, and Taylor diagram performance. Panel a shows a histogram of forecasting errors for Hybrid c G A ... More about this image found in Prediction error comparison with diverse machine learning models (family 1)...

Figure 7. Prediction error comparison with ML (family 2) Source: Authors’ own work Three panels compare forecasting error distributions, fitted error distributions, and Taylor diagram performance for prediction models. Panel a shows histograms of forecasting errors for Hybrid, E N, X G B,... More about this image found in Prediction error comparison with ML (family 2) Source: Authors’ own wo...

Figure 8. Prediction error comparison with return-based specification Source: Authors’ own work Three panels compare forecasting error distributions, fitted error distributions, and Taylor diagram performance for return forecasts. Panel a shows histograms of forecasting errors in W T I retu... More about this image found in Prediction error comparison with return-based specification Source: Auth...

Figure 1. DEMATEL-based ANP Method A process flowchart illustrating the integrated D E M A T E L and A N P methodology, beginning with expert assessment and progressing through influence matrix construction, network relationship mapping, supermatrix development, weighting, and stabilization. ... More about this image found in DEMATEL-based ANP Method A process flowchart illustrating the integrat...

Figure 2. Network relation map: (a) driver dimensions, (b) driver factors in economic dimension, (c) driver factors in social dimension, (d) driver factor in environmental dimension A network relation map showing causal relationships among Environmental, Economic, and Social dimensions based on... More about this image found in Network relation map: (a) driver dimensions, (b) driver factors in economic...

Figure 3. Network relation map: (a) barrier dimensions, (b) barrier factors in economic dimension, (c) barrier factors in social dimension, (d) barrier factor in environmental dimension A causal influence map displaying relationships among three economic driver factors labelled D E 1, D E 2, an... More about this image found in Network relation map: (a) barrier dimensions, (b) barrier factors in econom...

Figure 1. Serial mediation model showing the direct and indirect paths between variables A structural equation model shows relationships among A I D D I, G A C, G E O, G I, E P, and F P variables. The structural equation model presents relationships among 6 latent variables labelled A I D D I,... More about this image found in Serial mediation model showing the direct and indirect paths between variab...

Figure A1. Revised artificial intelligence device use acceptance (RAIDUA) model Source: RAIDUA structural model ( Salam et al., 2025 ) A conceptual model shows factors influencing acceptance and objection towards the use of A I devices through expectancy and emotions. The conceptual model illustrates relationships among Social Influence, Hedonic Motivation, Anthropomorphism, Usefulness Expectancy, Performance Expectancy, Emotions, Privacy Concerns, and responses towards the use of A I devices. Social Influence positively affects Usefulness Expectancy, with a coefficient of 0.572, and Performance Expectancy, with a coefficient of 0.576. Hedonic Motivation positively influences Usefulness Expectancy, with a coefficient of 0.486, and Performance Expectancy, with a coefficient of 0.421. Anthropomorphism positively affects Usefulness Expectancy, with a coefficient of 0.448, and Performance Expectancy, with a coefficient of 0.427. Usefulness Expectancy and Performance Expectancy both positively influence Emotions, with coefficients of 0.443 and 0.230, respectively. Emotions positively influence Willingness to Accept the Use of A I Devices, coefficient 0.694, and negatively influence Objection to the Use of A I Devices, coefficient minus 0.249. Privacy Concerns are linked with dashed moderating paths towards the relationships involving Emotions, Willingness to Accept the Use of A I Devices, and Objection to the Use of A I Devices, with coefficients minus 0.181, minus 0.115, and 0.103, respectively. The model also reports explained variance values, including R-squared 0.792 for Usefulness Expectancy, R-squared 0.718 for Performance Expectancy, R-squared 0.745 for Emotions, R-squared 0.764 for Willingness to Accept the Use of A I Devices, and R-squared 0.910 for Objection to the Use of A I Devices. More about this image found in Revised artificial intelligence device use acceptance (RAIDUA) model Sour...

Figure 1. Actors by scope of activity and territory Source: Own authorship A conceptual framework showing organisational coordination levels across national, regional, and local health and palliative care structures. The conceptual framework maps organisations according to Territorial Scope... More about this image found in Actors by scope of activity and territory Source: Own authorship A co...

Figure 2. Simple positions matrix actors × objectives (1MAO) Note: Values of 1, −1 and 0 represent agreement, disagreement and neutrality, respectively, regarding each strategic objective Source: Own authorship based on MACTOR software results A consensus matrix showing agreement across... More about this image found in Simple positions matrix actors × objectives (1MAO) Note: Values of 1, −1...