Background and Challenges
A large domestic supermarket has hundreds of stores across the country. Different stores have different sizes, different customer groups, and independent sales strategies. The supermarket group hopes to use machine learning technology to more accurately predict the sales of each store in advance to achieve precise, refined and intelligent operations.
Based on the daily sales of hundreds of stores, we need to analyze the trend, periodicity and irregularity of sales indicators changes over time, and predict the daily sales of each store in the future in order to help stores formulate reasonable and accurate financial plans and goals to achieve data-driven intelligent operations.
Difficulties and Key Success Factors
Using limited historical data to comprehensively analyze the impact of common and different factors from different stores on sales, and digging out more detailed rules from historical data are the key success factors to provide accurate predictions.
Through the 4paradigm Sage platform, a high-dimensional machine learning model is constructed to achieve detailed depictions of various factors in different stores, and then to make accurate predictions of future sales.
In addition, based on the Sage platform, the model has regular self-learning and uses the latest sales data to adjust itself to optimize the prediction of future sales indicators.
For the evaluation of historical data for half a year, the 4Paradigm solution with machine learning controls the relative error of the daily granularity sales forecast within 15% overall, which has greatly improved the existing expert rules.
At this stage, the 4Paradigm uses machine learning technology to analyze the trend, periodicity and irregularity of sales indicators change over time to achieve accurate prediction of sales indicators with different granularities, and will expand more practical offline application scenarios and committed to reducing business operating costs at multi-latitude and accurately predicting the company's development direction.