Figure 3
The diagram is divided into two horizontal sections. The left section is labeled “a) Hybrid M L with R F and E L Net”, and contains four subsections distinguished by numbered colored dashed borders: level 0 at the top includes “Trade data” and “Industrial production index data” flowing into “Replace missing value”, then “Normalize and extract features from data”, and “Split to train and validation set” leading to “Train or test data”. This leads to the next box in level 1, on the center left, labeled “Split to k folds”. Downward arrows from this box leads to parallel paths labeled “Training data” and “Test data” feeding “Random Forest” and “Elastic net”, respectively, which then merge and lead to “Evaluate each model”. A right arrow from “Evaluate each model” leads to Level 2, on the center right, with a box labeled “Combine results in hybrid model”, which leads rightward to the box labelled “Evaluate hybrid model”. A download arrow from “Split to train and validation set” in the level 0, leads downward to a box in level 2 labeled “Validation data”. A downward arrow from this box also leads to “Evaluate hybrid model”. “Evaluate hybrid model” leads downward to level 3 at the bottom with the final output stage labeled “Provide the final estimations”. A legend at the bottom indicates that level 0 represents “Data preparation”, level 1 represents “Training initial models”, level 2 represents “Combining in hybrid M L”, and level 3 represents “Final result”. The right section is labeled “b) Ensemble M L with stacked method”, and contains four subsections distinguished by numbered colored dashed borders: level 0 at the top includes “Trade data” and “Industrial production index data” flowing into “Replace missing value”, then “Normalize and extract features from data”, and “Split to train and validation set” leading to “Train or test data”. This leads to the next box in level 1, on the center left, labeled “Split to k folds”. Downward arrows from this box lead to parallel paths labeled “Training data” and “Test data” feeding “Random Forest” and “Elastic net”, respectively, which then merge and lead to “Evaluate each model”. A downward arrow from “Evaluate each model” leads to level 2 on the center left, with a box labeled “Stack the prediction from two models”, which leads rightward to the box labeled “Pass results to the Meta-model (Elastic net)”, in level 3 on the center right. A downward arrow from “Split to train and validation set” in level 0 leads downward to a box in level 3 labeled “Validation data”. A downward arrow from this box also leads to “Evaluate stacked model”. “Pass results to the Meta-model (Elastic net)” leads downward to “Evaluate stacked model”, and a downward arrow from this box goes to the bottom output stage labeled “Assess results and accuracy”. A legend at the bottom indicates that level 0 represents “Data preparation”, level 1 represents “Training base models”, level 2 represents “Stack the results from base”, and level 3 represents “Stacked M L evaluation”.Flowchart of the proposed ML SC models