Research On Real Estate Enterprise’s Financial Crisis Warning Model By Using Cost-sensitive Learning

Financial crisis warning model’s establishment is on an ideal hypothesis thattraining set is absolutely right, which means that there is financial crisis in financialcrisis enterprises and there is no potential financial crisis in financial healthenterprises. But when we establish financial crisis warning model, how could weguarantee the correctness of training set’s category if we use BP neural networksmethod during training and simulation. It is fatal significance for financial crisiswarning model based on the training set sample if there is wrong about the trainingset category, because this category mistake affects the accuracy of financial crisiswarning model, which is crucial for the financial crisis warning model’s success orfailure.Based on former scholar ’s research on financial crisis warning index system,the paper retains indexes extracted from the balance sheet, income statement andstatement of cash flows, such as, solvency, profitability, operating capacity, abilityto grow, ability to expand and index about cash flows and combines index aboutcooperate governance and audit opinion. The paper abandons these fuzzy indexeslike strategic management, staff relationship and life cycle and index which can becalculated from existing index like EVA. Meanwhile, the paper puts forward landreserve index to highlight financial management characteristics of real estatecompanies and policy restrictions index to highlight China’s real estate marketpolicy based on policy on finance, tax and deal registration.When the paper designs financial crisis warning process, BP neural networkscategory method is used to reflect the idea of cost-sensitive learning. After that, thepaper establishes financial crisis warning model based on cost-sensitive learning andBP neural networks. The specific progress of the model is as follows: firstly, thepaper standardizes41indexes of9aspects. Secondly, the samples are divided intotraining set and sample tested. Thirdly, the paper abandons the wrong samples toreflect the cost-sensitive learning idea after simulation and put the rest of sampletested back to the training set. Fourthly, the paper does the former work again until5results are completely right. This is financial crisis warning model based oncost-sensitive learning the paper establishes.