The RFM model: R (Recency) indicates how long the customer's last purchase was, and F (Frequency) represents the number of times the customer purchased in the most recent period, and M (monetary) represents the amount the customer purchased in the most recent period. The general raw data is 3 fields: Customer ID, purchase time, purchase amount, processed with data mining software, weighted (consider weight) get RFM score, sort score, output marketing list topn!
From @ Data Mining and data analysis
Here we operate the RFM model with IBM Modeler 14.1: (using data Mining to analyze RfM is a simple task because the software is very intelligent, or basically an automated RFM Analysis module is built in)
First: We read the data, assuming we have nearly 70,000 transaction data, including the customer ID, the transaction date and the volume; Generally speaking, in order to save storage space, the Time field is text type, so we first use the Fill field to turn date into a true date type field;
Because the RFM module for IBM Modeler14.1 has specific requirements for fields, we'll change first, remembering to select Always
This allows us to select the RFM transformation node to prepare R-proximate, F-frequency, m-Currency transaction data for RFM analysis; Select the RFM rollup node from the logging options
Note: At this time you can select a point in time: Specify the time or analysis time
Here, you'll find it's very simple, the node automatically converts the transaction dataset summary to the RFM analysis dataset: Proximate cause, frequency, and currency;
Now we need to split the three fields: select the RFM analysis node from the field options,
Specify three RFM fields proximate cause, frequency and currency, we need to decide the number of bins and weights, generally speaking: for different products and promotions, need to consider the different customer wishes, often the proximate weight is bigger, followed by patronage frequency; Here I have a simple processing: So the weights are the same, weighted = 10;
The RFM Analysis node field completes the base RFM analysis to get the proximate, frequency, and monetary score for each customer, and then calculates the weighted sum to get the RFM score.
The traditional RFM analysis is almost there, we need to sort out the customer RFM score and choose the 2000 people with the highest RFM score for promotion!
A real business campaign requires a customer response model that associates the RFM score with the customer's database and response data to create a targeted response model:
In general, assuming that the enterprise has 1 million member database and historical transaction data, you can randomly select 10,000 people for promotion before the data mining model is established, record the response of the 10,000 people, and then build the response model of RFM as an independent variable according to the 10,000 response data 0-1. Then according to the model of 1 million users to score customer response, choose 100,000 people to promote, can improve the ROI of the promotion;
The previous blog post is the use of IBM SPSS Statistics RFM Analysis, now we can see Data mining software analysis more convenient, more systematic, of course, more intuitive to operate!
Transferred from: http://shenhaolaoshi.blog.sohu.com/178101622.html
RFM model--Build a business campaign for database marketing! Go