Segmentaion for customers, to predict what they would but on their next visit according to their history
â–º InvoiceNo : Invoice number. Nominal, a 6-digit integral number uniquely assigned to each transaction. If this code starts with letter 'c', it indicates a cancellation.
â–º StockCode : Product (item) code. Nominal, a 5-digit integral number uniquely assigned to each distinct product.
â–º Description : Product (item) name. Nominal.
â–º Quantity : The quantities of each product (item) per transaction. Numeric.
â–º InvoiceDate : Invice Date and time. Numeric, the day and time when each transaction was generated.
â–º UnitPrice : Unit price. Numeric, Product price per unit in sterling.
â–º CustomerID : Customer number. Nominal, a 5-digit integral number uniquely assigned to each customer.
â–º Country : Country name. Nominal, the name of the country where each customer resides.
we cluster the products using tf-idf technique to get each term a score based on its importance in such product
and created matrix to be applied on K-Means Clustering algorithms
We add such product categories through the main dataframe, such that each Customer got a defined Category Some New Features were added to such Dataframe, such as:
- Number of visits
- total Cashe payed
- Min/ Max Cashe payed
- Average Cashe payed
After that we Begin the Clustering with K-Means Clustering algorithms which ended to be 11 customer categories
The final Dataframe will contains all information of each Customer Cluster Dimensions (11,13) ::
- 11 Clusters
- 13 Feature (
cluster,[count,sum,Avg,min,max], [6 product category])
Such Dataframe will be the Key for the followin Gerat Radar Visulization
Such fabulous chart indicates how Clusters are more to buy from such Category
- Customers of Cluster 0 are highly biased to buy products from 0 category
- Customers of Cluster 1 are also impressed with category 1 products
- Customers of Cluster 2 more to buy products from category 3 than any other category
models Used [Support Vector Machine, Logostic Regression, k-Nearest Neighbors, Decision Tree, Random Forest, Gradient Boosting, Ada Boost]
Which Ends with 87.38 % Prediction accuracy