IMPLEMENTATION OF K-NEAREST NEIGHBOR ALGORITHM FOR CREDIT SCORING ON PEER-TO-PEER (P2P) LENDING

• Main Author: Fikri Hafiya, Muhammad
• Format: Final Project
• Language: Indonesia
• Online Access: https://digilib.itb.ac.id/gdl/view/43718
• Institution: Institut Teknologi Bandung

One of the services in the financial industry is credit services, namely lending by one party to another party that is returned for a certain period of time with interest. Today, people can apply for credit not only to banks, but also to financial technology companies that provide peer-to-peer (P2P) lending that connects borrowers with investors without going through official financial institutions. P2P lending can provide access to finance to borrowers who may not have received approval from financial intermediaries. This concept can be a solution to the financial inclusion of the Indonesian people, for SMEs and to reach wider communities. P2P lending can open new segments of the economy, especially in communities that are not yet reached by financial institutions today. The growth trend of distribution of funds through P2P lending in Indonesia continued to increase during 2017-2018, but losses due to non-performing loans also increased. This requires a credit scoring system specifically made for P2P lending to reduce losses. In credit scoring, machine learning is often used for classification. The K-nearest neighbor (KNN) algorithm is a simple and suitable algorithm for credit scoring. To get optimal results, optimization is done by finding the best alternative parameter combinations between the value of the nearest neighbor (‘K’), the weighting of the neighbor, and the distance metric. After optimization, the best alternative parameter is the ‘K’ with the value of 41, with weighting depending on distance, and Canberra distance metric. This optimized model produces an accuracy value of 94.9%, precision 90.1%, recall 84.2%, f-measure 87.0%, specificity 97.6%, and AUC 90.9%.