The Synergy of Statistical and Fuzzy Logic Approaches in Mining Patterns from the Peer-to-Peer Lending Data

Abstract

Statistical measures, such as correlation, compute numeric values. However, it is not always the best option for domain experts. A promising way is to augment these measures linguistically. Therefore, the main objective of this work is the synergy of statistical and fuzzy logic approaches in mining and interpreting valuable information from financial lending data. The correlation reveals whether attributes are related while exhibiting relatively low computational costs. Fuzzy functional dependencies recognize the direction of influence but are demanding in terms of computational cost. Finally, linguistic summaries explore and interpret dependencies between the subdomains of the considered attributes. These two approaches are less influenced by a smaller vagueness in the data. In addition, the support for decision making validated by diverse approaches and explained from different points of view is more reliable. These approaches are integrated and applied to peer-to-peer (P2P) anonymized lending data consisting of 266,483 loans. Among other things, a significant correlation between loan amount and loan duration (r = 0.25) is explained further, indicating that the direction of influence is slightly stronger from loan duration to loan amount than the opposite case. At the same time, the dependency is very strong from low duration to low amount, but relatively weak from high duration to high amount. Finally, further research and application directions are outlined.