A fuzzy logic based evolutionary neural network for automotive residual value forecast

A fuzzy logic based evolutionary neural network system, for automotive residual value forecast, is proposed and tested on five different vehicle lines. An evolution selects the best variables as input and optimizes neural network architecture and parameters adaptively through training the historical data as time series. Fuzzy logic based evaluation function in the evolutionary algorithm obtains both a small training error for the training data and a small forecasting error for the testing data so that it avoids overfitting during the training and guarantees the neural network's forecasting ability on the future unseen data. The forecasting simulation results based on the five vehicle lines are also presented.