Multivariate Time Series Forecasting on Sales Using Recurrent Neural Network (Case Study: Aqiqah Almeera)

Abstract

Sales forecasting is a crucial component in business decision-making, particularly in inventory management and marketing strategies. Accurate sales predictions can help companies maintain stock balance, design effective promotions, and minimize the risk of losses. This study examines the application of Multivariate Time Series Forecasting using Recurrent Neural Networks (RNNs) to more accurately predict product sales. By considering multiple variables such as product price, inventory levels, promotional activities, and temporal features, this approach aims to capture complex and interrelated patterns in historical data. RNNs are chosen for their ability to handle sequential data and learn temporal relationships among variables, thereby improving prediction accuracy.

This research adopts a quantitative method with a causal-associative approach, utilizing secondary data from the company’s sales records over the past two years. The data is analyzed using various preprocessing techniques such as data normalization, feature encoding, and correlation analysis for optimal feature selection before being fed into the RNN model. The model is trained using specific validation techniques to prevent overfitting. Model performance is evaluated using MAE and RMSE metrics to measure prediction accuracy and reliability. The results of this study are expected to produce an accurate and practical sales forecasting system that can be implemented by business practitioners to support more efficient, data-driven, and well-targeted decision-making processes.