| Name: | Description: | Size: | Format: | |
|---|---|---|---|---|
| 1.86 MB | Adobe PDF |
Advisor(s)
Abstract(s)
O objetivo deste Trabalho Final de Mestrado (TFM) foi o da criação de uma métrica capaz de replicar o valor real da quota de mercado, através de dados transacionais extraídos do cartão cliente de um retalhista. O método utilizado fez uso de algoritmos de machine learning na identificação das top 15 categorias de loja, da análise de cluster na agregação dessas mesmas categorias e da análise de regressão na identificação dos fatores que afetam a quota de mercado de um determinado cluster. A partir deste último passo foi possível construir a métrica pretendida com base nos coeficientes das variáveis identificadas como significativas para o referido cluster: vendas brutas, número de transações, número de artigos disponíveis e percentagem de desconto. Amétrica resulta da soma ponderada pelos coeficientes destas variáveis transacionais. Os resultados mostram que com a métrica é possível monitorizar a quota de mercado por via da sua estimação interna, sem ter de depender de dados de quotas de mercado fornecidos por fonte externa.
The objective of this Master Final Assignment (MFA) was to create a metric capable of replicating the real value of a retail market share, through transactional data extracted from a retailer's customer card. The method used made use of machine learning algorithms to identify the top 15 store categories, cluster analysis to aggregate these same categories and regression analysis to identify the factors that affect the market share of a given cluster. From this last step, it was possible to build the desired metric based on the coefficients of the variables identified as significant for that cluster: gross sales, number of transactions, number of items available and discount percentage. The metric results from the sum weighted by the coefficients of these transactional variables. The results show that with the metric it is possible to monitor the market share through its internal estimation, without having to rely on market share data provided by an external source.
The objective of this Master Final Assignment (MFA) was to create a metric capable of replicating the real value of a retail market share, through transactional data extracted from a retailer's customer card. The method used made use of machine learning algorithms to identify the top 15 store categories, cluster analysis to aggregate these same categories and regression analysis to identify the factors that affect the market share of a given cluster. From this last step, it was possible to build the desired metric based on the coefficients of the variables identified as significant for that cluster: gross sales, number of transactions, number of items available and discount percentage. The metric results from the sum weighted by the coefficients of these transactional variables. The results show that with the metric it is possible to monitor the market share through its internal estimation, without having to rely on market share data provided by an external source.
Description
Keywords
Quota de mercado Dados transacionais Big data Machine learning Análise cluster Análise de regressão e métrica Retail market share Transactional data Cluster analysis Regression analysis and metric