This paper is Methodology research of Benefit-Cost Analysis (BCA) under machine learning-driven Classification Bayesian Neural Network. BCA is currently the predominant decision-making tool for corporate capital investment. Due to its inherent single-metric nature, BCA lacks complete quantifiability and responsiveness in terms of intangible benefits and ethics. In this paper the author presents how advanced AI scoring algorithm enables Multi-Attribute Decision Making (MADM) approach to dissolve monetization failure issue stemming from ethical concerns in Benefit-Cost Analysis (BCA). In the current context of increasing emphasis on Corporate Social Responsibility (CSR), ethical considerations have become vital for businesses. Therefore, it is advisable to utilize machine learning leveraging the business big data to train more sophisticated and stable multi-dimensional decision-making models, then integrate them into the framework of BCA, thereby evolving the BCA methodology to overcome its drawbacks. The three main problems associated with BCA discussed are: (1) intangible value is difficult to monetize, and (2) human life-related benefits should not be monetized due to the ethical principle. And (3) the normative and distributional problem of BCA that cannot take into account both the efficiency criterion and the equity criterion in one decision-making process. For the three problems of BCA, author introduces a Bayesian Neural Network Scoring algorithm based Analytic Hierarchy Process (AHP) to achieve Multi-Attribute Decision Making (MADM) and explains why it can solve the above three problems that are difficult for BCA to handle solely. Furthermore, arguing the limitations of such AHP-MADM methodology and puts forward the improvement methods including the Expected Utility Theory [1], and VIKOR method based on “Prospect Theory” [2, 3] which allow each organization to have its own unique decision preferences.