Análisis de la demanda de clientes bajo factores significativos – Cluster.

Abstract

This research analyzes customer demand by grouping significant factors across two branches of a supermarket located in El Oro Province, Ecuador. The study is motivated by the fact that consumers do not exhibit the same purchasing behavior regardless of purchase volume, purchase frequency, use of discounts and credit, customer type, and other commercial characteristics. Although total sales provide information about sales performance, they are not sufficient to understand the customer profiles that actually sustain and explain demand. Therefore, the main objective is to analyze demand through the grouping of relevant factors in order to understand consumer characteristics and identify the segment with the greatest impact on sales. The research follows a quantitative and analytical approach based on numerical information from commercial transactions at the two branches. The database includes sales, customers, products, discounts, credit, and transaction-related characteristics. Prior to applying the models, data preparation and cleaning were performed, including the identification of duplicate records, canceled transactions, data entry errors, outliers, and the organization and standardization of variables. This process ensured consistent information for statistical analysis and the application of machine learning techniques. An exploratory data analysis was also conducted to examine the distribution and behavior of the commercial variables. For customer segmentation, K-Means was applied as an unsupervised learning method for grouping observations with similar characteristics. The variables were standardized to prevent differences in scale from affecting the distances used by the algorithm. As a result, three customer groups were identified. Groups 0 and 1 showed relatively similar behavior, consisting mainly of individual customers with low sales volumes and product quantities. In contrast, Group 2 presented higher sales volumes and larger product quantities, greater use of credit, and a predominance of corporate customers, making it the most commercially intensive segment. Principal Component Analysis (PCA) was also used to graphically represent the segments and examine the differences among them. The results revealed three distinct groups, although Group 2 showed greater dispersion and some overlap with the other groups. Subsequently, multiple linear regression models were applied to the selected group, using sales volume as the dependent variable and variables such as selling price, product quantity, discount, credit, time factor, customer type, and sales location as explanatory variables to determine the factors most strongly associated with demand within the study segment. In conclusion, the results show that customers should not be analyzed as a single homogeneous group, since different demand behavior patterns can be identified. By combining K-Means and PCA, it is possible to identify commercial profiles and understand the factors associated with sales. This information can support the development of differentiated strategies for customer loyalty, promotions, credit management, inventory, and commercial planning, focusing on high-value customers and establishing specific incentives for customer segments with growth potential. Thus, analytical techniques and data-driven insights transform transactional data into a foundation for more accurate decision-making based on actual consumer behavior.