Peramalan Permintaan Donat dengan Pendekatan Machine Learning pada UKM Donat Kentang Syifa

Abstract

UKM Donat Kentang Syifa is a company engaged in the production of donuts. The company successfully produces approximately ± 4.000 – 5.000 donuts per day with 20 flavor variants. Donut sales at UKM Donat Kentang Syifa fluctuate every month. On the other hand, there are several leftover flavor variants of donuts that remain unsold, resulting in wastage. This is because the company still estimates the number of donuts to be produced manually, consequently unable to predict how many donuts should be produced to meet consumer demand. This research aims to forecast donut demand in 2024 using a machine learning approach. The methods used are the Naïve-Bayes algorithm and Prophet, starting from data collection, data preprocessing, model training, and model evaluation. The data used consists of historical daily donut sales data in 2022 and 2023, totaling 730 data points. This research produces a forecast of donut demand for the next 12 months with varying demand levels. Model evaluation results in a Mean Square Error (MSE) value of the Naïve-Bayes algorithm (428328.92), which is smaller than the Prophet algorithm (2717828e+07). This indicates that the Naïve-Bayes algorithm is more accurate in forecasting demand compared to the Prophet algorithm. The results of this research provide insights to the company in managing production accurately, thus avoiding stock shortages or surpluses