Implementation of the Naive Bayes Method for Sales Prediction at UD. Noer Berkah

Abstract

Sales prediction is essential for supporting production planning and inventory management in business operations. This study aimed to implement the Naïve Bayes classification method to predict sales status at UD. Noer Berkah using historical sales data. A total of 730 sales records collected from October 2024 to September 2025 were processed through data cleaning and preprocessing, resulting in 240 valid records. The dataset was divided into 192 training records and 48 testing records using an 80:20 hold-out method. The prediction model used month, work shift, meatball sales, and frozen food sales as input attributes, while sales status was used as the target class. Model performance was evaluated using a Confusion Matrix based on accuracy, precision, and recall. The model achieved an accuracy of 75%, with precision values of 85.71%, 66.67%, and 100%, and recall values of 70.59%, 90.91%, and 44.44% for the highly sold, sold, and unsold classes, respectively. These results indicate that the Naïve Bayes method provides satisfactory classification performance for identifying sales patterns from historical transaction data. The developed model can support production planning, inventory management, and business decision-making while reducing the risk of overstocking and stock shortages.