A Tool to predict mortgage default and recommend mortgage amount using convolution neural networks

Abstract

The mortgage sector is vital to the financial services industry and the Kenyan economy in general. In the period preceding March 2021, mortgage defaults surged by 48 percent to reach Sh70.5 billion in Kenya, signaling widespread distress within the real estate sector in the aftermath of economic challenges triggered by the Covid-19 pandemic. This surge was accompanied by a notable increase in property auctions. According to the latest data released by the Central Bank of Kenya (CBK), mortgages experienced the most substantial rise in non-performing loans (NPLs), soaring from Sh47.5 billion in March 2020. Overdue mortgages witnessed a staggering increase of Sh9.1 billion, equivalent to 14.8%, within the three-month period leading up to March, surpassing the default rates observed in other sectors such as manufacturing (3%), agriculture (10.7%), and personal loans (3%). As businesses adopt stringent cost-cutting measures to safeguard profits, mortgage holders find themselves grappling with financial strain in an economy marred by widespread job losses across various sectors since the onset of the Covid-19 pandemic in Kenya. Consequently, individuals who secured mortgages based on their employment income are now facing challenges in meeting their repayment obligations. The downturn in the real estate market poses significant challenges for property developers, who find themselves unable to offload units constructed using loans. This research led to the development of a tool that aids banks and other lending institutions in predicting the likelihood of a client defaulting on their mortgages using convolutional neural networks. The developed tool further recommends mortgage amount to the lenders to minimize the risk of defaulting. The developed model attained an impressive accuracy rate of 97.14%, surpassing the accuracy scores of Gradient Boosting and only slightly behind KNN model, which achieved 88.49% and 99.24%, respectively. The Agile Methodology was selected as the preferred approach owing to its emphasis on collaboration and facilitation of continuous improvement processes.

Keywords: non-performing loans, mortgage default, convolutional neural networks.