Data Survey for Elonen Oy Bakery: applying clustering to sales data analysis

Abstract

The background of this work was Elonen Oy Bakery's desire to better utilize collected sales and other data. The task was to investigate the available data and discuss with industry experts the best potential methods to use. The goal was to find a perspective that could be further developed and to create a machine learning model benefiting the business. Initially, various ideas were developed and presented to the company's representatives. Ultimately, the main topic became the application of clustering to analyze sales data. The primary goal was to cluster customers in the product group space and products in the area space, with other configurations also investigated. A thorough examination of K-Means and agglomerative clustering, their differences, and their application to Elonen's data was conducted. Skewness management of distributions and PCA and UMAP dimensionality reduction methods were also used. The result was a consistent clustering outcome for each dataset created from the sales data. While some data points may shift between clusters in different runs, the overall characterization of the clusters remains stable. Additionally, clusters were examined in relation to the natural categories of the data points. To improve usability, interactive scatter plots were created, making it easy to explore the data points and their relationships. It was found that multiple clustering configurations can be created from the same initial data, describing the data from different perspectives and providing insights into various matters of interest. In conclusion, clustering is particularly useful when someone unfamiliar with the industry wants to quickly find connections and patterns in the data, which would be slow and random to find using traditional data analysis methods.