Please use this identifier to cite or link to this item:
http://repository.ipb.ac.id/handle/123456789/166943| Title: | Model Prediksi Permintaan Produk Lapis Bogor ABC Menggunakan Algoritma LSTM dan Prophet di PT XYZ |
| Other Titles: | Lapis Bogor ABC Demand Prediction Model Using LSTM and Prophet Algorithms at PT XYZ |
| Authors: | Marimin RIZQILLAH, MUHAMMAD RASYID |
| Issue Date: | 2025 |
| Publisher: | IPB University |
| Abstract: | PT XYZ mengalami inefisiensi operasional akibat ketidakakuratan prediksi permintaan produk Lapis Bogor ABC yang hanya mencapai 80%, menyebabkan kerugian akibat overstock dan understock. Penelitian ini bertujuan untuk mengembangkan dan membandingkan model machine learning untuk meningkatkan akurasi prediksi dengan fokus pada algoritma Prophet dan Long Short-Term Memory (LSTM). Dengan menggunakan data penjualan historis dari Januari 2022 hingga Desember 2024, model dilatih dengan rekayasa fitur yang mencakup pola mingguan dan hari libur, serta dioptimalkan menggunakan genetic algorithm (GA). Hasil penelitian menunjukkan bahwa model Prophet yang dioptimalkan dengan GA secara signifikan lebih unggul, berhasil mencapai target akurasi 90% (MAPE 10%), sementara model LSTM menunjukkan kinerja yang tidak memuaskan. Temuan kunci mengungkapkan bahwa faktor prediktif paling dominan bukanlah penanda peristiwa liburan, melainkan data historis penjualan dari 7 dan 365 hari sebelumnya, yang menunjukkan bahwa efek peristiwa seperti Lebaran bersifat redundan dan telah terserap oleh komponen musiman tahunan. Implikasinya, model yang dikembangkan ini dapat menjadi alat bantu strategis untuk mengurangi kesalahan perencanaan dan berpotensi melakukan penghematan biaya, serta memberikan landasan pengambilan keputusan yang lebih objektif. PT XYZ experiences operational inefficiencies due to inaccurate demand forecasting for its product, which only achieves 80% accuracy, leading to losses from overstock and understock. This research aims to develop and compare machine learning models to improve forecasting accuracy, focusing on the Prophet and Long Short-Term Memory (LSTM) algorithms. Using historical sales data from January 2022 to December 2024, the models were trained with feature engineering that included weekly patterns and holidays, and optimized using a genetic algorithm (GA). The results show that the Prophet model optimized with GA was significantly superior, successfully achieving the 90% accuracy target (MAPE 10%), while the LSTM model performed unsatisfactorily. A key finding reveals that the most dominant predictive factors were not the holiday event markers, but the historical sales data from 7 and 365 days prior, indicating that the effects of events like Eid al-Fitr (Lebaran) were redundant and absorbed by the yearly seasonality component. The implication is that this developed model can serve as a strategic tool to reduce planning errors and potentially save costs, also providing a more objective basis for decision-making. |
| URI: | http://repository.ipb.ac.id/handle/123456789/166943 |
| Appears in Collections: | UT - Agroindustrial Technology |
Files in This Item:
| File | Description | Size | Format | |
|---|---|---|---|---|
| cover_F3401211109_89beda9354ed4406906d795ea35bb10d.pdf | Cover | 1.91 MB | Adobe PDF | View/Open |
| fulltext_F3401211109_4706deb7d08d4cc09df4bfb9183edadc.pdf Restricted Access | Fulltext | 3.01 MB | Adobe PDF | View/Open |
| lampiran_F3401211109_e2260d73b9a341eb985bfa02f0fa6ffc.pdf Restricted Access | Lampiran | 2.77 MB | Adobe PDF | View/Open |
Items in DSpace are protected by copyright, with all rights reserved, unless otherwise indicated.