RetailSales--Prediction

• The goal of a retail purchase prediction is to accurately forecast the demand for products and services to better manage inventory, anticipate customer needs, and maximize profits.
• By leveraging data-driven models and predictive analytics, retailers can accurately forecast future sales and make more informed business decisions.

Problem Statement:

• Exploratory data analysis to understand customer buying pattern
• Build a regression model to predict purchase amount of customer against various products

Tools used:

• Python, Pandas (data processing), Plotlyexpress,Sklearn

Data:

The dataset has 550,069 rows and 12 columns

Attributes:

Column ID	Column Name	Data type	Description	Masked
0	User_ID	int64	Unique Id of customer	False
1	Product_ID	object	Unique Id of product	False
2	Gender	object	Sex of customer	False
3	Age	object	Age of customer	False
4	Occupation	int64	Occupation code of customer	True
5	City_Category	object	City of customer	True
6	Stay_In_Current_City_Years	object	Number of years of stay in city	False
7	Marital_Status	int64	Marital status of customer	False
8	Product_Category_1	int64	Category of product	True
9	Product_Category_2	float64	Category of product	True
10	Product_Category_3	float64	Category of product	True
11	Purchase	int64	Purchase amount	False

Exploratory Data Analysis:

• There are 5891 users in the dataset and 3631 uique products
• 31% of Product_Category_2 and 69% of Product_Category_3 has missing values.
• Average amount spent by female customers is 8k and male is 9k
• Age group 26-35 has the highest total purchase across age groups
• Product category 0 has highest revenue of purchase and Product category 4 has highest number of purchases
• Unmarried customers in the age group of 26-35 have highest total purchase amount as compared to other customers
• Maximum purchases where customer have stayed in the city only for 1 year

Data Preprocessing:

• Product category 2 and 3 has missing values, we will use SimpleImputer to fill missing values with median values.
• Handle categorical columns Gender, Age, City_Category,Stay_In_Current_City_Years
• Drop UserID,Product_ID columns

Modelling

• Features(X): Gender, Age, Occupation,City_category, Stay_In_Current_City_Years,Marital_Status,Product_Category_1,Product_Category_2,Product_Category_3

• Label: Purchase

• Train test split 75% training and 25% test set

• Evaluate using RMSE and RMSLE metric baseline model using LinearRegression,DecisionTreeRegressor and RandomForestRegressor. RandomForestRegressor() had the lowest RMSE and RMSLE score

• Apply GridSearchCV to find the best parameter for RandomForestRegressor

• Product category_1 seem to have highest effect on purchase

• Surprisingly gender has the least effect on purchase