TRIVIANANDA, MARIA THERESIA DUMA (2021) COMPARATIVE DEMAND FORECAST USING LINEAR AND NON LINEAR MODEL TO OPTIMIZE OPERATIONAL PERFORMANCE IN FURNITURE RETAILER: A STUDY IN IKEA INDONESIA. Masters thesis, INSTITUT TEKNOLOGI SEPULUH NOPEMBER.
Preview |
Text
09211850086025-Master_Thesis.pdf - Accepted Version Download (3MB) | Preview |
Abstract
Penelitian ini dilakukan untuk mengetahui metode peramalan permintaan yang tepat untuk ritel perabot. Dengan kondisi permintaan yang tidak pasti dan rendahnya akurasi peramalan dibandingkan dengan aktual penjualan, maka ritel perabot seperti IKEA Indonesia membutuhkan evaluasi untuk metode peramalan permintaan yang sudah ada. Hal ini dilakukan untuk mengurangi resiko penimbunan maupun kekurangan stok barang yang disebabkan oleh perubahan tren permintaan pasar. Penelitian ini dilakukan dengan metode kuantitatif deskriptif dimana sumber penelitian merupakan data penjualan 2014-2018 yang akan dijadikan dasar peramalan tahun 2019 dan tahun 2020 ketika pandemi terjadi. Data penjualan tersebut diambil dari sistem IKEA Indonesia. Hasil dari peramalan akan dibandingkan dengan penjualan aktual tahun 2019-2020. Eksperimen dilakukan pada 14 produk area. Teknik sampling yang digunakan merupakan purposive sampling yang diambil dari beberapa pertimbangan seperti kelengkapan data, kategori barang, indeks kenaikan penjualan tiap tahun. Model yang digunakan adalah model yang memiliki performa akurat ketika diaplikasikan dalam kondisi krisis. Keempat metode yang digunakan adalah autoregressive integrated moving averages (ARIMA), seasonal autoregressive integrated moving averages (SARIMA), exponential smoothing (ETS) dan neural network (NN). Pengukuran ketepatan menggunakan 3 metode yaitu mean absolute percentage error (MAPE), mean absolute error (MAE) dan root mean squared error (RMSE). Jangka waktu peramalan dalam penelitian ini adalah 6, 12, 26, dan 52 minggu. Hasil penelitian memaparkan bahwa setiap produk area memiliki permodelan peramalan yang berbeda-beda untuk dapat mencapai akurasi yang tinggi. Model yang paling akurat untuk setiap produk area adalah ANN backpropagan menghasilkan peramalan dengan nilai error terendah ketika digunakan untuk meramalkan permintaan produk area toys for small children, duvet covers, system cabinet, bathroom accessories, pots and cooking, vases and bowls accessories. Sedangkan food containers, bathroom small furniture, dinnerware, dining chair dan floor lamp mencapai tingkat error minimum ketika permintaan produk area tersebut diramalkan menggunakan ARIMA, SARIMA, dan ANN backpropagan. Untuk produk area chest of drawers dan kitchen wall organizer mencapai keakurasian yang tinggi ketika diramalkan menggunakan SARIMA dan ANN backpropagan. Produk area terakhir adalah accessories for baby akurat ketika diramalkan menggunakan SARIMA, ANN dan ETS. Namun ketika model dengan tingkat keakurasian tinggi digunakan untuk meramalkan permintaan di masa pandemi, hasil menunjukkan 4 produk area memiliki tingkat akurasi yang rendah. Jangka melakukan peramalan berpengaruh besar terhadap tingkat akurasi dimana penelitian ini memaparkan bahwa jangka ideal untuk melakukan peramalan adalah 6 sampai 12 minggu. Begitu juga dengan metode pengukuran ketepatan, hasil menunjukkan bahwa MAPE ideal digunakan untuk membuat laporan untuk manajemen karena menunjukan presentase error pada suatu peramalan namun untuk tim operasional dan rantai pasok, RMSE dengan angka absolut lebih memberikan gambaran jelas untuk menyiapkan kapasitas dan tindakan yang harus diambil oleh tim penjualan apabila terjadi kehabisan atau kelebihan stok barang. Penelitian ini menjadi dasar perubahan model peramalan di IKEA Indonesia dan referensi kepada team sales untuk memperhatikan produk area tertentu yang mengalami perubahan permintaan di masa pandemi.
================================================================================================
This study was done to define the most suitable forecasting method for furniture retailer. The uncertainty condition and low forecast accuracy vs actual sales enforce furniture retailer like IKEA Indonesia to do an evaluation on the demand forecast method that they use. This needs to be done to avoid overstock and shortage due to shift of trend on demand from market. Quantitative – descriptive method was used in this study. The data source is 2014-2018 sales history that was used as the base of forecasting for 2019 and 2020 in pandemic situation. The actual sales data was taken from IKEA Indonesia system. The result of the forecast was compared to actual sales of 2019-2020. The simulation was done for 14 product areas. The samples were purposive sampling chosen based on come criteria such as data completion, product category, and growth indexes. The methods being used are the ones that robust for business all over the world during crisis situation. The four models were autoregressive integrated moving averages ARIMA, seasonal autoregressive integrated moving averages (SARIMA), exponential smoothing (ETS) and neural network (NN). The accuracy level of each method will be measured using 3 measures. Those three measures were mean absolute percentage error (MAPE), mean absolute error (MAE) and root mean squared error (RMSE). The lead of each forecasting was 6, 12 26 and 52 weeks to define the ideal length of forecasting for every PA. ANN backpropagan was the most accurate with lowest error model for toys for small children, duvet covers, system cabinet, bathroom accessories, pots and cooking, vases and bowls accessories. Meanwhile ARIMA, SARIMA and ANN backpropagan were resulting the minimum error when they were used to forecast food containers, bathroom small furniture, dinnerware, dining chair and floor lamp. For chest of drawers and kitchen wall organizer product area, the lowest error came from SARIMA and ANN backpropagan. The last one was accessories for baby with lowest error when they were forecasted using SARIMA, ANN backpropagan, and ETS. The result of this study showed that every product area in IKEA Indonesia had different method of forecasting and ideal lead of time to achieve the most accurate result were 6 to 12 weeks for all product areas. When the robust models were used to forecast 2020 (pandemic condition) the output of accuracy significantly decrease for 4 product areas. While to measure error MAPE was still suitable for management report. However, for operational and supply chain purpose RMSE gave clearer overview and guidance for action plan, capacity planning and sales steering since the number is absolute and the error weighting was calculated in every week. The purpose of this study was to rebuilt forecasting process in IKEA Indonesia and gave sales team an overview on product areas that having different demand during normal and pandemic condition.
| Item Type: | Thesis (Masters) |
|---|---|
| Uncontrolled Keywords: | Kata Kunci: Retail perabot, Metode peramalan, ARIMA, SARIMA, ETS, NN. ========================================================= ========================================================= Keywords: Furniture retailer, Forecasting method, SARIMA, ARIMA, ETS, NN |
| Subjects: | Q Science > QA Mathematics > QA276 Mathematical statistics. Time-series analysis. Failure time data analysis. Survival analysis (Biometry) |
| Divisions: | 61101-Magister Management Technology |
| Depositing User: | Maria Theresia Duma Triviananda |
| Date Deposited: | 19 Aug 2021 05:36 |
| Last Modified: | 19 Nov 2024 08:57 |
| URI: | http://repository.its.ac.id/id/eprint/87606 |
Actions (login required)
 |
View Item |