The instability of the financial system is likely to occur when particular types of loans surge rather than all types of loans surge at the same time. A preemptive policy response requires a monitoring system based on forecasts by different loan types. The purpose of this study is to forecast household loans bycategorizingintofourtypes:bankmortgageloan,bankcreditloan,non-bank mortgage loan, and non-bank credit loan. Given the fact that there are numerous determinants and forecasting models for household loans, and that the determinants differ depending on the type of household loans, this study sets out the density forecasting algorithm based on Bayesian Machine Learning. which consists of a variable learning process, a model learning process, and a forecasting combination process. We find bank mortgage loans are largely predicted by the loan rates, the volume of apartments to be moved in, and the number of apartment units to be sold. while the key determinants of bank credit loans are the employmentrateandJeon-sepriceindex.Ontheotherhand,thenon-bankmortgage loans are largely determined by the loan rates and the ratio of apartment sales prices relative to Jeon-se prices. The non-bank credit loans are also influencedbynotonlytheemploymentrateandtheJeon-sepriceindexbutalsostock returns.