[basic data mining technology] exploratory data analysis

1. Descriptive statistics （ Exploratory data analysis ）

Not all projects use complex methods , Descriptive statistics is a simple method .

Case study 1： Telecommunication company

Customers are the biggest assets of telecom companies , Customer behavior exists in the call record of the switch

Understanding customer behavior has become the trend of telecom companies ,

1. Help telecom companies find customers who need telephone lines and let them undertake several more telephone lines , Can bring greater business opportunities to the company

2. When will customers need additional connection lines

We have the data of the switch , Convert time seconds into start time and end time , How to sort time , Get the number of busy lines at each time point , I found that sometimes the line is full , It may cause the customer's incoming call to be busy , Lost business opportunities . As long as two more lines are installed, this problem can be solved .

Simple problems are solved in simple ways .

Case study 2：

I'm sure we've all seen it when shopping in the mall , A preferential plan for a good gift over 500 .

Especially on anniversaries or holidays , This kind of interval preferential scheme is the best marketing scheme , Once consumers hear that there will be a discount if there is a little difference in the amount , It is possible to buy a little more to get a discount .

Customer unit price is an indicator of marketing , Represents the total amount of customers' one-time consumption

Usually, the industry will sell through bundling , Complementary products are placed （ Flashlight battery ）, To increase the unit price , Increase performance , And it says 500 Send a gift ,1000 Send hundreds , These preferential activities can effectively increase the customer unit price , Generally, the salesperson will say 500,1000,1500, The intuitive way of multiple attracts customers to buy more , But is this really the best preferential plan ？ It is possible that this will lose a lot of performance in vain

This data is the total consumption amount of each customer of a clothing e-commerce ,

We calculate the per customer unit price per consumer

Altogether 21598 Consumer records ,ID： Customer number

Total_Spending The field represents the sum of the total consumption of a customer .

We use the total consumption amount as the horizontal axis , The number of customers is y Axis , Draw a histogram , You can see from the picture , The largest distribution of all customers is 150—200 The interval between is the most people , about 4000 Many people

Draw this density diagram , The dotted line represents the pricing strategy

If the optimization strategy is set at 500,150~200 A customer of yuan will feel that it will cost more than twice as much to get a discount , I definitely don't want to be a wrongdoer , If you have already purchased 450 Yuan is likely to buy extra . therefore 500 Yuan is not a suitable preferential strategy .

The appropriate pricing strategy should be based on consumers' consumption habits （200,300,500）

We can see that these peaks are the consumption habits of consumers

Therefore, we can translate several segmented amounts to improve performance .

So we can put 300,500,800 As pricing data .

（python drawing ）

Such simple descriptive statistics can effectively stratify customers , It can also be used to distinguish the member level of credit cards and other items . Again , No discount is allowed , Change to gifts （ Change according to user needs , For example, there was a hotel before , Their customers are business travelers , The money comes from the company , They prefer gifts and massages .）.

While increasing performance, it can also increase consumer satisfaction .

Statistics ： Take the previous magazine data as an example

Descriptive statistics （ The probability of buying various magazines ）： It can be seen that the most popular is house magazine ,（ The average ）,comic0.081

Prediction accuracy .Naïve predictions Basic standards , If the accuracy is the standard , that naïve predictions Is the basic probability .

Descriptive statistics ： The average value distribution of five magazines （ Customer differentiation ）：

and naïve prediction The comparison shows the relationship .

2. Visual computing （ Exploratory data analysis ）

Distribution of people who buy various magazines

Age distribution of purchasing magazines .（ Very uniform , It can be compared with different magazines .）

Age distribution of buying Auto magazines （ A little concentrated ）

Age is helpful in predicting whether to buy Auto magazines .

Age distribution of sports magazines （ Age has little effect on it .）

Age and income whether to buy music magazine scatter chart （ Light dots represent purchases , Black dots mean not to buy ）

Obviously, people who buy music tend to have low income and low age . Age and income should help predict the purchase of music magazines

summary ： Simple questions can be analyzed through exploratory data . Descriptive statistics can be combined with visual analysis .