Prediksi Penjualan Mobil menggunakan Convolutional Neural Network-Long Short Term Memory dengan Bayesian Optimization

Abstract

Vehicle sales data has complex characteristics, such as seasonal patterns, long-term trends, and irregular fluctuations, necessitating a prediction method capable of accurately capturing these patterns. This study aims to analyze the effect of Bayesian Optimization on the error rate of the Convolutional Neural NetworkLong Short Term Memory (CNN-LSTM) prediction model in predicting vehicle sales. This research method using CNN-LSTM was tested in two scenarios: without optimization and with Bayesian Optimization as the best hyperparameter search technique. The CNN-LSTM model was tested in two scenarios: without optimization and with Bayesian Optimization as the best hyperparameter search technique. Performance evaluation was conducted using RMSE, MAE, and MAPE metrics, as well as visual analysis of prediction graphs and actual data. The results show that Bayesian Optimization is generally able to reduce the prediction error rate for most brands,. The optimized model produces more stable predictions and better aligns with actual patterns. Bayesian Optimization has a positive impact on improving the accuracy of the CNN-LSTM model in vehicle sales forecasting and can be recommended as an effective optimization approach.