Research On Financial Risks Prediction: An Improved Random Subspace Method For Using Multi-source Heterogeneous Data

With the deepening of the reform of China's financial industry,the Internet financial model represented by P2 P lending,crowdsourcing,third-party payment and etc.has received increasing prevalence,which promotes enterprise's development and financial market's efficiency.In the meanwhile,factors hastened by the open Internet environment such as the fierce market competition and the fast-changing market environment have impelled financial market players to be confronted with severe challenges from financial risks.Financial risks prediction can provide such players with effective and prompt warning signals as well as decision support so that their financial losses can be avoided or reduced to some extents.Therefore,how to predict financial risks with respect to the Internet environment as accurate as possible has raised a common concern of both academic research and industry applications in recent years.At present,numerous researchers have studied on financial risks and proposed a series of feasible financial risks prediction methods.However,most of them are constructed based on a single data source,resulting in limited prediction accuracy.Besides that,the prediction performance achieved by these methods commonly varies drastically from different data scenes.Recently,some researchers begin to predict financial risks utilizing multi-source data,by which the improvements on prediction effects are somewhat not obvious.We attribute this to that they simply superpose the obtained data of multiple sources for use but neglecting the inherent data characteristic such as data's different sources,heterogeneous structures,and redundancy.Hence,novel financial risks prediction methods are constructed in this study to conduct adapted integration on multi-source heterogeneous data,so as to make full use of predicion information and maximally improve the prediction performance.First,this study systematically summarizes the research with regard to financial risks prediction.In the light of their deficiencies,we construct a financial risks prediction method of weightfused adaptive integration-based Random Subspace(i.e.,WFAIB_RS)for listed companies and a financial risks prediction method of two steps adaptive integration-based Random Subspace(i.e.,TSAIB_RS)for individual borrowers from Peer-to-Peer(P2P)lending market,respectively.Then,we verify the proposed methods WFAI_RS and TSAIB_RS by applying them into the crawled and collected datasets with respect to Chinese listed company and individual borrowers from Chinese P2 P lending market respectively.The experimental results show that both of the two methods have achieved remarkable prediction performance in their respective application scenary,which demonstrates their feasibility and effectiveness for financial risks prediction.On the one hand,this study systematically combs the theories redarding financial risks prediction and thoroughly analyzes the impacts of different data sources,different data structures,and different data panels on financial risks prediction,which provides a theoretical basis for using multi-source heterogeneous data to predict financial risks.On the other hand,this study innovates to focus on financial risks of market's players from perspectives of company-level and individual-level,and proposes novel financial risks prediction methods that enables multi-source heterogeneous data to be adaptively integrated.This,in fact,has provided a guidance for applying multi-source heterogeneous data into the field of financial risks manegement in the big data era,and has also provided new possibilities to further breakthrough the accuracy level of existing financial risks prediction.