|
所在平台: Udemy |
课程主页: https://www.udemy.com/course/store-segmentation/
课程评论：没有评论
课程名称：利用数据科学进行零售店铺细分 课程概述：本课程指导您如何应用机器学习和数据科学技术，从原始数据中构建商店细分，以生成易于理解且可供利益相关者采取行动的细分。课程基于在零售公司实施的真实项目（由于保密原因使用合成数据），遵循数据科学生命周期中的关键步骤。 我们首先明确定义业务问题，并识别相关变量，包括客户人口统计、购物行为、部门级贡献、运营绩效、商店规模、市级经济指标和天气数据。接着，您将探索常见的数据源和提取方法（从数据仓库如BigQuery到API、网页抓取和Google Sheets）。 接下来，我们深入数据清理、预处理和特征工程，并通过关联矩阵、分布图和箱线图进行探索性分析。我们应用数据变换技术，如温莎化、Yeo-Johnson变换和标准化，之后执行主成分分析（PCA）以探索潜在结构并指导细分过程。 在建模阶段，我们专注于寻找最稳定的聚类解决方案，使用Jaccard相似性评估随机状态之间的一致性。我们利用肘部法则评估最佳聚类数量，并通过轮廓系数（Silhouette score）评估聚类质量。 为了描述所得细分，我们采用一种受SAS Miner启发的剖面技术。我们利用决策树识别每个细分的最具区分性的特征，然后可视化分布以比较每个细分与整体人口之间的差异。这使我们能够根据关键差异撰写简单且方便利益相关者理解的描述。 最后，我们总结所有内容并展示结果，为零售环境中基于数据的决策提供支持。
This course guides you through applying machine learning and data science techniques to build a store segmentation from raw data in order to generate actionable, easy-to-understand segments for stakeholders. Based on a real-world project implemented in a retail company (with synthetic data due to confidentiality), the course follows key steps in the data science lifecycle.We begin by defining the business problem and identifying relevant variables, including customer demographics, shopping behavior, section-level contributions, operational performance, store size, city-level economic indicators, and weather data. You'll then explore common data sources and extraction methods (ranging from data warehouses like BigQuery to APIs, web scraping, and Google Sheets).Next, we dive into data cleaning, preprocessing, and feature engineering, followed by exploratory analysis using correlation matrices, distribution plots, and boxplots. We apply data transformations such as winsorization, Yeo-Johnson, and standardization before running a PCA to explore latent structure and guide the segmentation process.For modeling, we focus on finding the most stable clustering solution, using Jaccard similarity to evaluate consistency across random states. We evaluate the optimal number of clusters with the Elbow method and assess quality of the clustering using Silhouette score.To describe the resulting segments, we adapt a profiling technique inspired by SAS Miner. We use decision trees to identify the most distinguishing features per segment, then visualize distributions to compare each segment against the overall population. This allows us to craft simple, stakeholder-friendly descriptions based on key deviations.Finally, we wrap everything up with a presentation of results, ready to support data-driven decision-making in a retail context.
