Machine Learning Used Car Price Prediction with Random Forest Regressor Model

Abstract

The pandemic has stopped all activities, including the decline in the level of the world economy, this will reduce people's purchasing power. For those who want to buy a car, a used car can be the second option. In some used car sales showrooms, the main task is to determine price predictions based on historical data during previous transactions. The determinants of car prices are heavily influenced by several attributes in the car, for example: type of fuel, for example, km traveled and so on, this is what causes the price prediction process to take a long time. One of the roles in Macine Learning is being able to learn from previous transaction data and this will be a model that can be used to provide used car price predictions. Car price prediction is included in regression, which is looking for a strong relationship from the influence of variable X (predictor) to variable Y (target). In the prediction, of course, the reality data will not be right with the predicted data, for that in the measurement the model will look for the lowest error rate. The experimental results on the test data using the Random Forest Regressor model resulted in MAE = 1.006 and RMSE = 1.452 while the coefficient of determinant R2 = 0.89. And in the previous study with KNN [18], it produced an error rate of MAE = 2.01 and RMSE = 4.01 and the coefficient of determinant R2 = 0.85. While the comparison model uses Linear Regression, Ridge, Decision Tree and Gradient Boosting. Not all Machine Learning models are suitable for all data, for that it is necessary to choose the right machine learning model by experimenting with several models. And the lowest error level (MAE and RMSE) will be determined. The error values ​​for MAE and RMSE which are close to zero are close to the predicted value close to the actual value. On the other hand, if the error rate is very high, the prediction value is very far from the actual value.