Advanced Demand Forecasting and Pricing in Moroccan Auto Industry: A CNN-LSTM-Attention and Reinforcement Learning Approach

This paper presents a novel forecasting model that combines Convolutional Neural Network-Long Short-Term Memory with an Attention mechanism (CNN-LSTM-Attention) and Reinforcement Learning (RL) to enhance impressive utilization in forecasting accuracy on demand for pricing optimization within auto parts industry in Morocco. The performance of the model was determined using Mean Squared Error (MSE) and R-squared (R2) as evaluation measures for prediction of demand accuracy while “Average Reward per Episode” was applied to measure how well pricing strategy works. The Application Programming Interface (API) utilized CNN-LSTM parameterization for direct spatial computation from images, Recurrent Neural Network (RNN) like sequential information processing and attended timestamps, which resulted in error rates below 1% while producing an R2 of around 90%. Regarding those metrics, it outperforms normal LSTMs or convolutional long short-term memory networks. By contrast, empirically Deep Q-Network (DQN) was more compatible than conventional Q-Learning when applied towards evolving pricing strategies over time leading up greater rewards and consistency amidst market volatilities. Market trend response was cores dependable, which resulted in significant readiness in supply chain operations. The present research work gives an example where smart predictive analysis can provide a comparative advantage.