<![CDATA[Abstract: Sales forecasting is a crucial aspect of supermarket operations, as it supports inventory management, production planning, and strategic decision-making. Sales data typically exhibit complex patterns such as trends, seasonality, and fluctuations, requiring modeling methods capable of handling nonlinear time-series characteristics. This study employs the Long Short-Term Memory (LSTM) model, an advanced form of Recurrent Neural Network (RNN) designed to capture long-term dependencies and overcome the vanishing gradient problem. The secondary dataset was obtained from the Kaggle platform, consisting of 20 features and a total of 1,000 records. The LSTM model was constructed using 50 neurons in the LSTM layer and a single dense output layer. Training was conducted for 100 epochs using the Adam optimizer and Mean Squared Error (MSE) as the loss function. The training process showed a consistent decrease in loss, reaching approximately 0.0193, while evaluation using Root Mean Squared Error (RMSE) indicated that the model effectively learned historical patterns. Visualization of predictions on the test dataset demonstrated that the model successfully followed sales trends, although it was less responsive to extreme fluctuations. Overall, the LSTM model proved effective for daily sales forecasting and can serve as a valuable tool for operational planning in supermarkets. Keywords: LSTM, Sales Forecasting, Time Series, Deep Learning, RMSE, Supermarket. Abstrak: Peramalan penjualan merupakan aspek penting dalam operasional supermarket karena berperan besar dalam pengelolaan inventaris, perencanaan produksi, serta pengambilan keputusan strategis. Data penjualan umumnya memiliki pola tren, musiman, dan fluktuasi yang kompleks sehingga memerlukan metode pemodelan yang mampu menangani karakteristik deret waktu nonlinear. Penelitian ini menggunakan model Long Short-Term Memory (LSTM), sebuah pengembangan Recurrent Neural Network (RNN) yang efektif dalam menangkap dependensi jangka panjang dan mengatasi masalah vanishing gradient. Data sekunder diperoleh dari platform Kaggle dengan 20 fitur dan total 1.000 record. Model LSTM dibangun menggunakan 50 unit neuron pada lapisan LSTM dan satu lapisan dense sebagai output. Model dilatih selama 100 epoch menggunakan optimizer Adam dan fungsi loss MSE. Hasil pelatihan menunjukkan penurunan loss yang stabil hingga mencapai nilai sekitar 0,0193, sedangkan evaluasi menggunakan Root Mean Squared Error (RMSE) menunjukkan bahwa model mampu mempelajari pola historis dengan baik. Visualisasi prediksi pada data pengujian memperlihatkan bahwa model mampu mengikuti tren pergerakan penjualan meskipun masih kurang responsif terhadap fluktuasi ekstrem. Secara keseluruhan, model LSTM terbukti efektif dalam memprediksi penjualan harian dan dapat digunakan sebagai dasar pengambilan keputusan dalam perencanaan operasional supermarket. Kata kunci: LSTM, Peramalan Penjualan, Deret Waktu, Deep Learning, RMSE, Supermarket.]]>
