Perancangan Model Peramalan Permintaan Bolu Lapis Menggunakan Machine Learning untuk Mendukung Proses Perencanaan Produksi

Abstract

Peramalan permintaan berperan penting dalam mendukung perencanaan produksi karena memengaruhi penentuan jumlah produksi, kebutuhan bahan baku, dan kapasitas produksi. Proses peramalan di PT XYZ masih menggunakan pendekatan berbasis pertimbangan subjektif (judgement) sehingga diperlukan metode berbasis data untuk mendukung pengambilan keputusan. Proyek ini bertujuan menganalisis karakteristik permintaan produk Bolu Lapis varian Black Forest, mengembangkan mekanisme disagregasi hasil peramalan bulanan menjadi mingguan, mengevaluasi kinerja algoritma XGBoost, LightGBM, CatBoost, dan Support Vector Regression (SVR) dengan metode hyperparameter tuning, serta merancang prototype sistem peramalan berbasis web. Data yang digunakan berupa data penjualan bulanan periode Januari 2022 hingga Februari 2025. Tahapan proyek meliputi preprocessing data, analisis eksploratif data, feature engineering, pelatihan model, optimasi hyperparameter, evaluasi model, disagregasi dan pengembangan sistem menggunakan metode System Development Life Cycle (SDLC) model Prototype. Hasil analisis menunjukkan tingkat variasi permintaan yang tinggi dengan nilai Coefficient of Variation sebesar 88,9%. Model terbaik diperoleh menggunakan CatBoost yang dioptimasi dengan Optuna yang menghasilkan Mean Absolute Percentage Error (MAPE) sebesar 8,89%, Mean Absolute Error (MAE) sebesar 153,9, Root Mean Squared Error (RMSE) sebesar 247,7, dan akurasi sebesar 91,11%. Prototype sistem yang dikembangkan mampu mengimplementasikan model peramalan untuk menghasilkan peramalan permintaan bulanan dan mingguan serta berpotensi mendukung penyusunan rencana produksi berbasis data.

Demand forecasting plays an important role in supporting production planning because it influences production quantity determination, raw material requirements, and production capacity. The forecasting process at PT XYZ still relies on a subjective judgement-based approach, therefore, a data-driven method is required to support decision-making. This project aims to analyze the demand characteristics of the Black Forest variant of Bolu Lapis products, develop a disaggregation mechanism from monthly forecasting results into weekly periods, evaluate the performance of XGBoost, LightGBM, CatBoost, and Support Vector Regression (SVR) algorithm using hyperparameter tuning methods, and design a web-based forecasting system prototype. The data used in this project consisted of monthly sales data from January 2022 to February 2025. The project stages included data preprocessing, exploratory data analysis, feature engineering, model training, hyperparameter optimization, model evaluation, demand disaggregation, and system development using the System Development Cycle (SDLC) with the Prototype model. The results showed that demand variation was high, with a coefficient of variation value of 88.9%. The best model was obtained using CatBoost optimized with Optuna, which resulted in a Mean Absolute Percentage Error (MAPE) of 8.89%, a Mean Absolute Error (MAE) of 153.9, a Root Mean Squared Error (RMSE) of 247.7, and an accuracy of 91.11%. The developed prototype was able to implement the forecasting model to generate monthly and weekly demand forecasts and has the potential to support data-driven production planning.