Four 1 million-level apparel sales datasets（dresses, jeans, sweaters, and sweatshirts）

These datasets were developed through a collaboration between Zhejiang Sci-Tech University and Hangzhou Zhiyi Technology Co., Ltd., encompassing a four-year span from January 2019 to October 2023. We comprehensively document daily sales records of dresses, jeans, sweatshirts, and sweaters that maintained a sales volume exceeding 50 pieces and continued to sell for over 100 days. The dress dataset witnessed a maximum surge to 6.1 million, more than 12 times the smooth value at the surge point, while the jeans dataset recorded a peak of 7.3 million, over 18 times the smooth value. Similarly, the sweatshirt and sweater datasets exhibited surges of 13 and 16 times the smooth values, respectively. Such extreme volatility in the datasets poses a challenge for conventional algorithmic models to learn mutation data effectively without compromising linear fitting accuracy.

本数据集由浙江理工大学与杭州智易科技有限公司（Hangzhou Zhiyi Technology Co., Ltd.）联合研发，数据覆盖周期为2019年1月至2023年10月，跨度长达四年。本次数据集完整收录了连衣裙、牛仔裤、运动卫衣及针织毛衣四类服饰的每日销售记录，筛选标准为单品单日销量超50件且持续在售周期超过100天。其中连衣裙数据集的销量峰值达610万件，峰值时刻的波动幅度为对应平滑值的12倍以上；牛仔裤数据集的销量峰值为730万件，峰值波动幅度达对应平滑值的18倍以上。同理，运动卫衣与针织毛衣数据集的销量峰值分别为对应平滑值的13倍与16倍。该类数据集存在极强的波动特性，这对传统算法模型提出了挑战：需在不牺牲线性拟合精度的前提下，有效学习突变数据的分布规律。