ased for such corroborative research, thesuperior performance of

ased on the tests executed, the CNN based approach is capable of outperforming the Chow-Liu algorithm. While confidence in this specific result is high, it must be caveated; performancehas been adequately proven in the narrow, synthetic network specific, problem space but thisdoes not infer performance in generalised settings. Further discussion of specific test metricsgathered is provided within the subsequent sections.5.1 Accuracy, F1, Matthew’s Coefficient and ROC AUCThe experiments’ results demonstrated the CNN approach’s superiority to the Chow-Liu al-gorithm when applied to synthetic BNs generated in the same fashion as those used to trainthe CNN. This superiority was observed in accuracy, F1, ROC AUC and Matthew’s coefficientscores.Evidence within literature suggests that the Matthews Coefficient is a more robust measurementof binary classifications than ROC AUC and F1 scores (Powers 2011). Given the greatest deltawithin performance was observed with the Matthew’s coefficient this lends further credence tothe notion that the CNN method is capable of greater efficacy than the Chow-Liu algorithm.However as author bias is likely to be present in the search for such corroborative research, thesuperior performance of this measure was not examined in any depth.Unfortunately, the observed improvements in accuracy, F1 score etc. show a gradual declineas the BN becomes larger and hence more complex. Indeed this trend appears to regress to-wards the baseline accuracy i.e. no better than an informed guess, presuming knowledge of theapproximate edge density is known. This is in contrast to the Chow-Liu results with outputsthat become more accurate as more complex networks are proposed. However this increasedaccuracy remains below the baseline accuracy, and also appears to be converging (almost mir-roring the decline observed within the CNN prediction accuracy). Unfortunately owing to timeconstraints more complex networks were not explored, but would potentially garner a morecomprehensive view of this observed phenomena.5.2 Run TimeThe experiments also demonstrated the CNN approach’s superior execution times. However, aswith the prior metrics, this improved execution time does show a declining trend as the networkstructures become more complex, and may, in fact, be overtaken by the Chow Liu algorithm inlarger networks. For example while a 4 node network is completed 0.07s quicker by the CNNon average this drops to just 0.02s for a 10 node network.When algorithmic time complexity is considered alone the neural network approach should out-perform the Chow Liu algorithm given the Chow Liu algorithm is of time complexity O(n2lognBased on the tests executed, the CNN based approach is capable of outperforming the Chow-Liu algorithm. While confidence in this specific result is high, it must be caveated; performancehas been adequately proven in the narrow, synthetic network specific, problem space but thisdoes not infer performance in generalised settings. Further discussion of specific test metricsgathered is provided within the subsequent sections.5.1 Accuracy, F1, Matthew’s Coefficient and ROC AUCThe experiments’ results demonstrated the CNN approach’s superiority to the Chow-Liu al-gorithm when applied to synthetic BNs generated in the same fashion as those used to trainthe CNN. This superiority was observed in accuracy, F1, ROC AUC and Matthew’s coefficientscores.Evidence within literature suggests that the Matthews Coefficient is a more robust measurementof binary classifications than ROC AUC and F1 scores (Powers 2011). Given the greatest deltawithin performance was observed with the Matthew’s coefficient this lends further credence tothe notion that the CNN method is capable of greater efficacy than the Chow-Liu algorithm.However as author bias is likely to be present in the search for such corroborative research, thesuperior performance of this measure was not examined in any depth.Unfortunately, the observed improvements in accuracy, F1 score etc. show a gradual declineas the BN becomes larger and hence more complex. Indeed this trend appears to regress to-wards the baseline accuracy i.e. no better than an informed guess, presuming knowledge of theapproximate edge density is known. This is in contrast to the Chow-Liu results with outputsthat become more accurate as more complex networks are proposed. However this increasedaccuracy remains below the baseline accuracy, and also appears to be converging (almost mir-roring the decline observed within the CNN prediction accuracy). Unfortunately owing to timeconstraints more complex networks were not explored, but would potentially garner a morecomprehensive view of this observed phenomena.5.2 Run TimeThe experiments also demonstrated the CNN approach’s superior execution times. However, aswith the prior metrics, this improved execution time does show a declining trend as the networkstructures become more complex, and may, in fact, be overtaken by the Chow Liu algorithm inlarger networks. For example while a 4 node network is completed 0.07s quicker by the CNNon average this drops to just 0.02s for a 10 node network.When algorithmic time complexity is considered alone the neural network approach should out-perform the Chow Liu algorithm given the Chow Liu algorithm is of time complexity O(n2lognBased on the tests executed, the CNN based approach is capable of outperforming the Chow-Liu algorithm. While confidence in this specific result is high, it must be caveated; performancehas been adequately proven in the narrow, synthetic network specific, problem space but thisdoes not infer performance in generalised settings. Further discussion of specific test metricsgathered is provided within the subsequent sections.5.1 Accuracy, F1, Matthew’s Coefficient and ROC AUCThe experiments’ results demonstrated the CNN approach’s superiority to the Chow-Liu al-gorithm when applied to synthetic BNs generated in the same fashion as those used to trainthe CNN. This superiority was observed in accuracy, F1, ROC AUC and Matthew’s coefficientscores.Evidence within literature suggests that the Matthews Coefficient is a more robust measurementof binary classifications than ROC AUC and F1 scores (Powers 2011). Given the greatest deltawithin performance was observed with the Matthew’s coefficient this lends further credence tothe notion that the CNN method is capable of greater efficacy than the Chow-Liu algorithm.However as author bias is likely to be present in the search for such corroborative research, thesuperior performance of this measure was not examined in any depth.Unfortunately, the observed improvements in accuracy, F1 score etc. show a gradual declineas the BN becomes larger and hence more complex. Indeed this trend appears to regress to-wards the baseline accuracy i.e. no better than an informed guess, presuming knowledge of theapproximate edge density is known. This is in contrast to the Chow-Liu results with outputsthat become more accurate as more complex networks are proposed. However this increasedaccuracy remains below the baseline accuracy, and also appears to be converging (almost mir-roring the decline observed within the CNN prediction accuracy). Unfortunately owing to timeconstraints more complex networks were not explored, but would potentially garner a morecomprehensive view of this observed phenomena.5.2 Run TimeThe experiments also demonstrated the CNN approach’s superior execution times. However, aswith the prior metrics, this improved execution time does show a declining trend as the networkstructures become more complex, and may, in fact, be overtaken by the Chow Liu algorithm inlarger networks. For example while a 4 node network is completed 0.07s quicker by the CNNon average this drops to just 0.02s for a 10 node network.When algorithmic time complexity is considered alone the neural network approach should out-perform the Chow Liu algorithm given the Chow Liu algorithm is of time complexity O(n2lognBased on the tests executed, the CNN based approach is capable of outperforming the Chow-Liu algorithm. While confidence in this specific result is high, it must be caveated; performancehas been adequately proven in the narrow, synthetic network specific, problem space but thisdoes not infer performance in generalised settings. Further discussion of specific test metricsgathered is provided within the subsequent sections.5.1 Accuracy, F1, Matthew’s Coefficient and ROC AUCThe experiments’ results demonstrated the CNN approach’s superiority to the Chow-Liu al-gorithm when applied to synthetic BNs generated in the same fashion as those used to trainthe CNN. This superiority was observed in accuracy, F1, ROC AUC and Matthew’s coefficientscores.Evidence within literature suggests that the Matthews Coefficient is a more robust measurementof binary classifications than ROC AUC and F1 scores (Powers 2011). Given the greatest deltawithin performance was observed with the Matthew’s coefficient this lends further credence tothe notion that the CNN method is capable of greater efficacy than the Chow-Liu algorithm.However as author bias is likely to be present in the search for such corroborative research, thesuperior performance of this measure was not examined in any depth.Unfortunately, the observed improvements in accuracy, F1 score etc. show a gradual declineas the BN becomes larger and hence more complex. Indeed this trend appears to regress to-wards the baseline accuracy i.e. no better than an informed guess, presuming knowledge of theapproximate edge density is known. This is in contrast to the Chow-Liu results with outputsthat become more accurate as more complex networks are proposed. However this increasedaccuracy remains below the baseline accuracy, and also appears to be converging (almost mir-roring the decline observed within the CNN prediction accuracy). Unfortunately owing to timeconstraints more complex networks were not explored, but would potentially garner a morecomprehensive view of this observed phenomena.5.2 Run TimeThe experiments also demonstrated the CNN approach’s superior execution times. However, aswith the prior metrics, this improved execution time does show a declining trend as the networkstructures become more complex, and may, in fact, be overtaken by the Chow Liu algorithm inlarger networks. For example while a 4 node network is completed 0.07s quicker by the CNNon average this drops to just 0.02s for a 10 node network.When algorithmic time complexity is considered alone the neural network approach should out-perform the Chow Liu algorithm given the Chow Liu algorithm is of time complexity O(n2lognBased on the tests executed, the CNN based approach is capable of outperforming the Chow-Liu algorithm. While confidence in this specific result is high, it must be caveated; performancehas been adequately proven in the narrow, synthetic network specific, problem space but thisdoes not infer performance in generalised settings. Further discussion of specific test metricsgathered is provided within the subsequent sections.5.1 Accuracy, F1, Matthew’s Coefficient and ROC AUCThe experiments’ results demonstrated the CNN approach’s superiority to the Chow-Liu al-gorithm when applied to synthetic BNs generated in the same fashion as those used to trainthe CNN. This superiority was observed in accuracy, F1, ROC AUC and Matthew’s coefficientscores.Evidence within literature suggests that the Matthews Coefficient is a more robust measurementof binary classifications than ROC AUC and F1 scores (Powers 2011). Given the greatest deltawithin performance was observed with the Matthew’s coefficient this lends further credence tothe notion that the CNN method is capable of greater efficacy than the Chow-Liu algorithm.However as author bias is likely to be present in the search for such corroborative research, thesuperior performance of this measure was not examined in any depth.Unfortunately, the observed improvements in accuracy, F1 score etc. show a gradual declineas the BN becomes larger and hence more complex. Indeed this trend appears to regress to-wards the baseline accuracy i.e. no better than an informed guess, presuming knowledge of theapproximate edge density is known. This is in contrast to the Chow-Liu results with outputsthat become more accurate as more complex networks are proposed. However this increasedaccuracy remains below the baseline accuracy, and also appears to be converging (almost mir-roring the decline observed within the CNN prediction accuracy). Unfortunately owing to timeconstraints more complex networks were not explored, but would potentially garner a morecomprehensive view of this observed phenomena.5.2 Run TimeThe experiments also demonstrated the CNN approach’s superior execution times. However, aswith the prior metrics, this improved execution time does show a declining trend as the networkstructures become more complex, and may, in fact, be overtaken by the Chow Liu algorithm inlarger networks. For example while a 4 node network is completed 0.07s quicker by the CNNon average this drops to just 0.02s for a 10 node network.When algorithmic time complexity is considered alone the neural network approach should out-perform the Chow Liu algorithm given the Chow Liu algorithm is of time complexity O(n2lognBased on the tests executed, the CNN based approach is capable of outperforming the Chow-Liu algorithm.  While confidence in this specific result is high, it must be caveated; performancehas been adequately proven in the narrow, synthetic network specific, problem space but thisdoes not infer performance in generalised settings.  Further discussion of specific test metricsgathered is provided within the subsequent sections.5.1    Accuracy, F1, Matthew’s Coefficient and ROC AUCThe  experiments’ results  demonstrated the  CNN approach’s  superiority to  the Chow-Liu  al-gorithm when applied to synthetic BNs generated in the same fashion as those used to trainthe CNN. This superiority was observed in accuracy, F1, ROC AUC and Matthew’s coefficientscores.Evidence within literature suggests that the Matthews Coefficient is a more robust measurementof binary classifications than ROC AUC and F1 scores (Powers 2011).  Given the greatest deltawithin performance was observed with the Matthew’s coefficient this lends further credence tothe notion that the CNN method is capable of greater efficacy than the Chow-Liu algorithm.However as author bias is likely to be present in the search for such corroborative research, thesuperior performance of this measure was not examined in any depth.Unfortunately, the observed improvements in accuracy, F1 score etc.  show a gradual declineas the BN becomes larger and hence more complex.  Indeed this trend appears to regress to-wards the baseline accuracy i.e.  no better than an informed guess, presuming knowledge of theapproximate edge density is known.  This is in contrast to the Chow-Liu results with outputsthat become more accurate as more complex networks are proposed.  However this increasedaccuracy remains below the baseline accuracy, and also appears to be converging (almost mir-roring the decline observed within the CNN prediction accuracy).  Unfortunately owing to timeconstraints  more  complex  networks  were  not  explored,  but  would  potentially  garner  a  morecomprehensive view of this observed phenomena.5.2    Run TimeThe experiments also demonstrated the CNN approach’s superior execution times.  However, aswith the prior metrics, this improved execution time does show a declining trend as the networkstructures become more complex, and may, in fact, be overtaken by the Chow Liu algorithm inlarger networks.  For example while a 4 node network is completed 0.07s quicker by the CNNon average this drops to just 0.02s for a 10 node network.When algorithmic time complexity is considered alone the neural network approach should out-perform the Chow Liu algorithm given the Chow Liu algorithm is of time complexity O(n2logn