<p>In this study, we present a hybrid model of combining Long Short-Term Memory (LSTM) with Multi-Layer Perceptron (MLP) to predict weekly retail sales. While sales data are time dependent data, several features are not necessarily have the same characteristics. LSTMs have limitations in capturing patterns or events that are not related to the time sequence (time-independent/non-temporal data). By combining LSTM and MLP, we hope MLP can complement the capture of events or patterns that are not well captured by LSTM. We evaluate our method on Walmart Recruiting Store Sales Forecasting data. To measure the performance of the hybrid model, five metrics were used: Mean Absolute Error (MAE), Root Mean Squared Error (RMSE), Mean Squared Error (MSE), Mean Absolute Percentage Error (MAPE), and the coefficient of determination (R²). This LSTM-MLP hybrid model showed better performance with MAE of 2.39, RMSE of 3.39, MSE of 14.07, MAPE of 12.39, and R² approaching 1. This model outperformed other architectures such as the standalone MLP, GRU, and LSTM in terms of predicting weekly retail sales on the dataset tested in this study. This performance indicates that the hybrid LSTM-MLP method is effective in capturing sales patterns. The proposed method shows better results especially for sales with larger volumes, which have relatively lower number of samples compared to that of sales with smaller number of sales. This may be the advantages of the proposed methods over transformers models.</p>