OPTIMIZING OPERATIONS OF LEBANESE STEEL COMPANY: FOZ TRADING USING DEMAND FORECASTING AND OPTIMAL ORDERING POLICY

Abstract

This thesis examines the various industry-used methods in time series forecasting for the Lebanese tool steel company: FOZ Trading. Previous sales data of FOZ Trading showed signs of intermittent demand in all of the categories examined. The limitations of usual forecasting methods in the case of intermittent demand, such as simple exponential smoothing and simple moving average, has prompted the use of intermittent demand-specific approaches along with several basic and traditional forecasting methods. Basic forecasting methods such as last period demand and simple moving average are benchmarked against traditional and alternate forecasting methods such as Box-Jenkins, Croston, Croston TSB and the simple exponential smoothing. Moving block bootstrapping and circular block bootstrapping were also applied on top of Croston TSB and simple moving average in order to check if they would perform better. Locally weighted linear regression and gaussian process regression were tested and evaluated for suitability as part of the machine learning methods. Additionally, temporal aggregation was applied to reduce the intermittency aspect of the data so that basic and traditional forecasting methods would perform better. As part of the evaluation process, bias correction and prediction intervals were generated to improve the performance and usability of the forecasts. All the methods were implemented using python and then tested and validated using the walk forward optimization. The best methods to use, based on the root mean squared error and the mean absolute error, were selected and then applied in the order up-to model to optimize the inventory level of the company. The whole demand forecasting and optimal ordering process was automated using python and Jupyter Notebook. The final application was provided to the company to use in their demand planning process. This allowed FOZ Trading to adjust their stocks and orders according to the forecasted demand and in line with the recommendations of the optimal ordering policy thus optimizing the inventory level and reducing cost in the short and long term.