Applying Big Data Algorithms For Sales Data Stored In Sap Hana

An increasing number of small and medium enterprises operate and supervise their financial, logistics, production, human resource and other activities with ERP (Enterprise Resource Planning) IT systems, which are capable of managing these processes in a uniform framework. It is our research objective to explore analytical and prediction methods which can provide solutions to various financial and logistics problems that are faced by nowadays’ enterprises. In order to achieve this objective, we analysed a real database of an enterprise using SAP. This database covers the period between January 2013 – October 2016 and contains 7 million sales records of 15,674 different products. These products also include currently inactive items, however, they can still be important from the aspect of data analysis in relation to the examined period. The data structure was created using the recently introduced SAP HANA (High Performance Analytic Appliance) database management system which revolutionised data storage with its in-memory and column-oriented features. This new technology makes it possible to execute various transactions more effectively and more quickly. Following the preparation of the schematics of data needed for processing and the calculation of calculation fields, we used the tools provided by SAP Predictive Analytics which was introduced to the market in 2015. After filtering sales data for 15 quarters, we used k-means clustering for each period. After preparing and examining the clusters, we made observations which make it easier to perform stock management, logistics and pricing activities in the future, thereby contributing to the long-term increase of enterprise profit. Clustering was also performed in R programming language, which enabled us to illustrate the clustering results, i.e., each sales record in 3D, colouring them based on their associated cluster label. After inspecting these graphic outputs, it was concluded that certain products should be withdrawn from the market, while others should be either developed or their stock level increased. We used regression analysis on the cluster centroids to predict the movement of each cluster mainly in terms of time. As a result, we provided an estimation as to the direction that the products belonging to the sales records in each cluster will fluctuate to in accordance with the coordinates of the cluster centroids, thereby making recommendations related to the management of each item and group of items. As a next step, we performed a global F test for regression analysis to examine the correctness of our model. As a conclusion, we reject the null hypothesis which stated that our model is basically invalid.