Author(s): Carreira, André Neves de Almeida Roque
Date: 2017
Persistent ID: http://hdl.handle.net/10362/22165
Origin: Repositório Institucional da UNL
Subject(s): Storage allocation; Big Data; Phantom stock; Machine Learning; Algorithm
Author(s): Carreira, André Neves de Almeida Roque
Date: 2017
Persistent ID: http://hdl.handle.net/10362/22165
Origin: Repositório Institucional da UNL
Subject(s): Storage allocation; Big Data; Phantom stock; Machine Learning; Algorithm
Dissertation presented as the partial requirement for obtaining a Master's degree in Information Management, specialization in Knowledge Management and Business Intelligence
As we observe a rise of competitive pressure in retail business, major players control the market using promotions to attract and increase the fidelity of customers. These promotions cause a significant decrease in profit margin hence forcing a stronger focus on logistic processes efficiency. Forced to make use of stochastic tools in order to better allocate resources for the future events, those responsible for assortment strategy frequently choose time series based algorithms and simple extrapolation of historical data, under the assumption that these events can be considered continuous, smooth and possibly periodic. While computationally light, these algorithms are subject to greater uncertainty due to the simplistic approach. Meanwhile, the explosive growth of information and availability of data brought by improved automatic collection systems allow new and more complex approaches.These tackle the high dimensionality problem, focused on retrieving knowledge from potentially rich sources of information. The work developed in this thesis aims to develop a comprehensive and scalable solution using machine learning algorithms to forecast daily sales of articles in a retail store, under the influence of discounts, as to support logistic storage allocation operations. This is done with the purpose of decreasing costs related to stock warehousing while simultaneously decreasing stock-outs as they directly influence client satisfaction with the brand. The development of a successful automatic modelling system would simultaneously allow retailers to optimize their promotional schedules based on the expected results of different simulations. Using real data from one of the biggest retailers in Portugal, this project’s falls into the definition of Big Data due to extensive historical databases which cannot be simultaneously processed. The presence of discrepancies between registered stock and physical availability - Phantom stock - will be considered as well as relevant external events which affect the sales.