Deep Learning for Enterprise Sales Prediction: Haressing CNN-LSTM-Attention Model

This study introduces a novel CNN-LSTM-Attention model to address limitations in traditional LSTM architectures, particularly in scenarios impacted by small weights. Leveraging Convolutional Neural Networks and a multi-head attention mechanism with Dropout, the model optimizes the LSTM component for improved predictive capabilities. Experimental results on enterprise revenue prediction demonstrate a substantial decrease in RMSE and MAPE, indicating the model's superior accuracy compared to traditional LSTM and LSTM-Attention models. The proposed approach presents a promising avenue for advancing time series forecasting in scenarios where nuanced dependencies affect predictive outcomes.