Seite - 38 - in Intelligent Environments 2019 - Workshop Proceedings of the 15th International Conference on Intelligent Environments

Figure1. WorkflowofPrediction. duction.The input is structured andunstructureddata. First, features are extracted from original unstructured data. The feature extraction methods include frequency method, TF-IDF,mutual informationmethodN-gram,wordembeddingandempiricalmanualde- sign. Then, the features extracted from text, togetherwith structured data, are fed into differentmachinelearningmodels todoprediction.Theoutput iscapitaladequacylevels. Table1. TheAdequacy IndicatorsofBankCapital. Warning Adequacy Adequacy Color Level Indicators 1.Majorcounterpartieshaveabundant capital Green Normal 2.Thesupplyofovernight capital is sufficient andovernightprice financedbynon-bank institutions isnothigh 3.Amplemoneysuppliedbypolicybacks toX-REPO 1.Majorcounterpartieshave1/3 shrinkageofmoneysupplyor stop lending Yellow Tight 2.Theovernight capital limitedsupplyorfinancedovernight capital withhighprices fornon-bank institutions 3.Policybanks reduce theovernight capital supplyonX-REPO 1.Majorcounterpartieshave2/3 shrinkageofmoneysupplyor stop lending Orange Very 2.Theovernight capital supply is scarceorfinanced long term Tight capitalwithhighprices fornon-bank institutions 3.Policybanksdonot supply theovernight capital onX-REPO Red Extremely 1.Majorcounterparties stop lending Tight 2.Banksdonot lendmoney tonon-bank institutions Y.Duetal. /Predicting the InterbankCapitalAdequacyLevelBasedonFinancialDataAnalysis38