An Adaptive algorithm to predict bad debts during COVID-19 Pandemic

Loan recovery during the COVID-19 pandemic is anxious. Automated decision-making would boost the identification of bad debts while issuing loans. The objective of the proposed work is thus to design and implement an adaptive algorithm, which will be used to predict bad debts. Machine learning is an artificial intelligence technology, which gives systems the ability to automatically learn and improve from experience without explicit programming. The adaptive algorithm proposed is deterministic, uses two parameters known as neighborhood distance and minimum support threshold value for the risk profile, and can be very useful in predicting bad debts. It produces overlapped as well as non-overlapped clusters. This algorithm can detect the outliers with the help of an adaptive threshold value for the object's risk profile attribute. Objects with a moderately high or high value of risk profile attribute may emerge as outliers, and these outliers can be known as bad debts. The clusters generated are labeled as paid fully, not paid fully, and not paid. It can also generate clusters of different sizes. The proposed adaptive deterministic algorithm clusters the dataset without knowing the number of clusters. Many clusters are generated using this algorithm, but the parameter risk profile minimum threshold value prunes the clusters being formed. This proposed adaptive algorithm is tested using real and artificial data sets and shows 83% accuracy in bad debt prediction.