This study proposes a novel technique by extending balanced scorecards with risk management considerations (i.e., risk metrics and insolvency risk) for corporate operating performance assessment and then establishes a fusion mechanism that incorporates hybrid filter-wrapper subset selection (HFW), random vector functional-link network (RVFLN), and ant colony optimization (ACO) for operating performance forecasting. The study executes HFW, which preserves the advantages of wrapper approaches, but prevents paying its tremendous computational cost, in order to determine the essential features for forecasting model construction. With the merits of rapid learning speed and no extra inherent parameters needed to be tuned, RVFLN helps establish the forecasting model. However, RVFLN has demonstrated that its superior forecasting performance comes with the challenge of being unable to represent the inherent decision logic for humans to comprehend. To cope with this task, the study conducts ACO so as to extract the inherent knowledge from RVFLN and represents it in human-readable format. If the extracted knowledge is not comprehensive for decision makers, then they will not be able to interpret and verify it. In this circumstance, the decision makers probably will not trust enough the extracted knowledge and be prone to making unreliable judgments more easily. The introduced mechanism herein is examined by real cases and poses superior forecasting quality under numerous examinations. It is a promising alternative for corporate operating performance forecasting.
