What are the data analysis requirements in business scenarios
• Descriptive ability, ability to describe with data (business indicators, statistical methods)
• Forecasting capabilities, forecasting user profiles, sales forecasting, inventory forecasting, predictive maintenance
• Analytical ability, understanding of causes, explanatory features
• Automation, 7*24 automatic operation, continuous action (not one-time)
Kraljic Model (Data-driven procurement positioning)
Strategic Items
Procurement items that are critical to the production process often have high supply risks due to scarce supplies or transportation difficulties
Position of buyer and seller: balance of power, high interdependence
Procurement strategy recommendation: strategic alliance, close connection, early involvement of suppliers, co-creation, and full consideration of vertical integration, focus on long-term value
Bottleneck Items
Procurement items that can only be supplied by a specific supplier, are difficult to transport and have low financial impact
Buyer and seller status: the seller is active and interdependent.
Procurement strategy recommendations: quantity insurance contracts, vendor management inventory, securing additional inventory, sourcing potential suppliers
RFM user value model
RFM indicators:
Recency: the interval between the last consumption
Frequency a number of purchases made within a period of time
Monetary The amount of money spent over a period of time such as a year
For example, on August 22nd, user A purchased A product in the store, and the last purchase was on August 15th. Please ask Recency=?
The larger each indicator is, the higher the user value is. Use three indicators as the XYZ coordinate axis and divide the space into eight parts for analysis
Intelligent supply chain
Supply Chain data Exploration
1) Check whether there are missing fields => Data completion
2) How does the correlation between these features => present the thermogram
3) Explore Sales volume (corresponding to Sales per Customer)
According to different Market, Order Region
By different Category names
Trends in different time dimensions (year, month, week, hour)
How does Product Price correlate with Sales per Customer
RFM is used to manage users hierarchically
Predict fraud
To predict the fraudulent Order, Order Status=’SUSPECTED_FRAUD’
Forecast Sales performance
3) Explore Sales volume (corresponding to Sales per Customer)
According to different Market, Order Region
By different Category names
Trends in different time dimensions (year, month, week, hour)
How does Product Price correlate with Sales per Customer
GBDT vs. XGBoost
GBDT is a machine learning algorithm and XGBoost is an engineering implementation of the algorithm
XGBoost adds regularization terms to control the complexity of the model, which helps to prevent overfitting and improve the generalization ability of the model
GBDT only uses the information of the first derivative of the cost function in model training, XGBoost carries out the second-order Taylor expansion of the cost function, and can use the first and second derivatives at the same time
Whereas traditional GBDT uses all the data in each iteration, XGBoost adopts a strategy similar to random forest and supports data sampling
Traditional GBDT is not designed to process missing values, XGBoost can automatically learn the missing value processing strategy