Analysis and Visualization of Kare Kambing Sales Forecasting Using Time Series-Based Forecasting and Machine Learning

Abstract

The Kare Kambing MSME possesses a substantial amount of historical sales data, yet its utilization as a basis for business decision-making remains suboptimal. Production planning is currently based on estimates, creating a risk of either overproduction or underproduction. This study aims to analyze sales patterns, compare various forecasting methods, and visualize the analysis results via a Business Intelligence dashboard to support business decision-making. The study utilizes historical sales data from the Kare Kambing MSME covering the period from 2020 to February 2026. The research stages include data preprocessing, exploratory data analysis (EDA), forecasting model development, model evaluation, sales forecasting, and dashboard creation using Tableau. The models compared are Moving Average, Exponential Smoothing, ARIMA, Linear Regression, and Random Forest. Model evaluation was conducted using MAE, RMSE, and SMAPE metrics, with data split into 80% for training and 20% for testing. The results indicate fluctuating sales patterns, characterized by day-to-day variations and relatively stable seasonal trends. Based on the evaluation, the Linear Regression model emerged as the best performer, achieving an MAE of 61.93, RMSE of 77.82, and SMAPE of 7.25. This model was subsequently used to forecast sales for the period of March 2026 through December 2027, yielding a total predicted volume of 17,623 portions. Finally, the forecast results were visualized in a Tableau dashboard to facilitate sales monitoring, model evaluation, and the structured formulation of raw material inventory recommendations.