Please use this identifier to cite or link to this item:
|Title:||Demand forecasting on time series with external correlation using deep learning techniques||Authors:||Teo, Boon Shuan||Keywords:||Engineering::Computer science and engineering||Issue Date:||2021||Publisher:||Nanyang Technological University||Source:||Teo, B. S. (2021). Demand forecasting on time series with external correlation using deep learning techniques. Final Year Project (FYP), Nanyang Technological University, Singapore. https://hdl.handle.net/10356/153859||Abstract:||Demand Forecasting is undoubtedly the most crucial step for any organizations dealing with Supply Chain - in fact, it is the first step for planners and decision-makers in the industry. This step determines the estimated demand for the organization’s goods or services in the immediate future, and sets the precedent for the organization’s supply side to increase or decrease production to meet said demand. It is imperative for an organization to have a reasonably accurate level of forecasting, as over or under forecasting would fail the organization in terms of optimizing operational costs - through minimizing costs and maximizing gains, and therefore Profits and Losses. While Demand Forecasting has been around since the 1980s lead by Spyros Madridakis, and has been heavily researched since, it is inconclusive between methods of traditional ML and Deep Learning as to which would yield better results, as the results vary greatly from dataset to dataset. Moreover, in most cases, many predictive time series algorithms fail to capture anomalies in their prediction, due to their training solely on historical data. In this project, we proposed a solution, to introduce external data in training the prediction algorithms. The rationale is that they would contain up to date information of the time series trends, and hence when included in the training phase, the models would be able to capture and foresee anomalous trends as in the external data, as opposed to relying solely on historical data of the target variable. We will do so by implementing: Baseline model: LSTM (Univariate) Experimental model: LSTM( (Multivariate) - which will include time series features from Google Trends We managed to achieve a total of 31/87 SKUs outperform in LSTM Multivariate against LSTM Univariate. Mean RMSE Percentage Change in performance is +30.4180528189771% across all 87 SKUs (increase in error). Mean RMSE Percentage Change in performance is -17.1403337006036% across 31 outperforming SKUs (decrease in error).||URI:||https://hdl.handle.net/10356/153859||Fulltext Permission:||restricted||Fulltext Availability:||With Fulltext|
|Appears in Collections:||SCSE Student Reports (FYP/IA/PA/PI)|
Updated on Jan 21, 2022
Updated on Jan 21, 2022
Items in DR-NTU are protected by copyright, with all rights reserved, unless otherwise indicated.