针对服装订单销售弹性较大、易出现大量库存积压的问题,提出一种基于周期基准孤立森林的异常监测算法.该算法在孤立森林的基础上加入了数据分解法,对数据进行周期性的处理.首先,对服装订单数据进行数据预处理,将负值和缺失值进行均值的填充;其次使用STL分解法对预处理后的数据进行数据分解,剔除数据中的季节项和趋势项,保留数据残差;最后,用孤立森林对分解后的数据进行异常数据识别.实验结果表明,所提算法相较于孤立森林算法准确率提高了6.11个百分点,受试者操作曲线下面积(AUC)提高了8.16个百分点.该算法适用于含有季节性和周期性的数据异常检测,能够为服装产量提供决策支持.
To address the problem that clothing orders and sales are highly elastic and susceptible to significant overstock,an anomaly detection algorithm based on isolated forests with periodic reference was presented.In this algorithm,the data decomposition method was added on the basis of isolated forest,and the data was processed periodically.Firstly,data preprocessing was carried out on the clothing order data to fill the negative and missing values with the mean value;then data decomposition was carried out on the preprocessed data by the STL decomposition method to remove the seasonal and trend terms in the data and retain the data residuals;lastly,anomaly data identification was carried out on the decomposed data by the isolated forest.Experimental results reveal that compared to the basic isolated forest algorithm,the proposed algorithm achieves a notable improvement of 6.11 percentage points in accuracy and an impressive 8.16 percentage points increase in AUC(Area Under the receiver operating characteristic Curve).This algorithm is well-suited for detecting data anomalies exhibiting seasonality and periodic patterns,offering valuable decision support for clothing production.