Predictive Analytics on Used Car Prices Using Business Intelligence of Bayesian Networks for Sales Risk Reduction

Used car prices are normally estimated at the time of initial purchase when the car is freshly purchased. Due to the shortage of cars from car manufacturers during the pandemic periods, secondhand car dealers have gained an undue advantage by withholding some information about cars advertised online. Researchers have tried to develop models to address this issue. There has not been enough research done in the area of uncertainty (partial observability) about used car prices. This research develops three transparent probabilistic models: a generalized model for all used cars, a specialized model for luxury cars, and a specialized model for non-luxury cars. It is intended to determine which models best predict used car prices. Publicly available car datasets were sourced, cleaned, and processed to build the models. Experimental evaluations reveal that the generalized model predicts prices with an accuracy of 80.19%, the specialized BN model for non-luxury cars predicts prices with an accuracy of 80.13% while the specialized BN model for luxury cars predicts prices with an accuracy of 83.5%. The results suggest that a specialized model performs better when predicting luxury used car prices. However, with non-luxury used cars, any of the two models predict prices satisfactorily. This research is aimed at helping used car buyers get a fair idea of expected selling prices, reducing risks using lower and upper bound prices. A new car buyer can also apply this model to estimate how much their car will be worth after a few years, which could aid in reaching sound business decisions.