• Hackers Realm

Black Friday Sales Prediction Analysis using Python | Regression | Machine Learning Project Tutorial

Black Friday Sales Prediction is a regression problem where we have to analyze and predict the sales of an product in the retail store based on various aspects of the dataset. The objective is to build a predictive model and discover the sales of each product.


In this project tutorial, we analyze and predict the sales during Black Friday, and display the results through plot graphs and different prediction models.



You can watch the video-based tutorial with step by step explanation down below


Dataset Information


This dataset comprises of sales transactions captured at a retail store. It’s a classic dataset to explore and expand your feature engineering skills and day to day understanding from multiple shopping experiences. This is a regression problem. The dataset has 550,069 rows and 12 columns.


Problem: Predict purchase amount



Attributes:

  • Masked attributes hide the data information.


Download the Dataset here



Import modules


import pandas as pd
import numpy as np
import seaborn as sns
import matplotlib.pyplot as plt
import warnings
%matplotlib inline

warnings.filterwarnings('ignore')
  • pandas - used to perform data manipulation and analysis

  • numpy - used to perform a wide variety of mathematical operations on arrays

  • matplotlib - used for data visualization and graphical plotting

  • seaborn - built on top of matplotlib with similar functionalities

  • %matplotlib - to enable the inline plotting.

  • warnings - to manipulate warnings details

  • filterwarnings('ignore') is to ignore the warnings thrown by the modules (gives clean results)



Loading the dataset


df = pd.read_csv('train.csv')
df.head()
  • Some columns have null values, those values must be replaced for a relevant value for further processing.


Let us see the statistical information of the attributes

# statistical info
df.describe()
  • Statistical information of the data

  • Product_Category_2 and Product_Category_3 have lower number of samples than Product_Category_1, both could be sub categories.



Let us see the data type information of the attributes

# datatype info
df.info()
  • We have categorical as well as numerical attributes which we will process separately.

  • Product_Category_1 data type is different from Product_Category_2 and Product_Category_3, that won't affect the process or the result.



# find unique values
df.apply(lambda x: len(x.unique()))
  • Attributes containing many unique values are of numerical type. The remaining attributes are of categorical type.



Exploratory Data Analysis


# distplot for purchase
plt.style.use('fivethirtyeight')
plt.figure(figsize=(13, 7))
sns.distplot(df['Purchase'], bins=25)
  • First part of the graph has a normal distribution and later forming some peaks in the graph

  • Evaluating the whole graph, it has a normal distribution



# distribution of numeric variables
sns.countplot(df['Gender'])
  • Many buyers are male while the minority are female.

  • Difference is due to the categories on sale during Black Friday, evaluating a particular category may change the count between genders.



sns.countplot(df['Age'])
  • There are 7 categories defined to classify the age of the buyers



sns.countplot(df['Marital_Status'])
  • Majority of the buyers are single



sns.countplot(df['Occupation'])
  • Display of the occupation of the buyers

  • Occupation 8 has extremely low count compared with the others; it can be ignored for the calculation since it won't affect much the result.



sns.countplot(df['Product_Category_1'])
  • Majority of the products are in category 1, 5 and 8.

  • The low no. categories can be combined into a single category to greatly reduce the complexity of the problem.



sns.countplot(df['Product_Category_2'])
  • Categories are in float values

  • Categories 2, 8, 14 to 16 are higher compared with the others.



sns.countplot(df['Product_Category_3'])
  • Categories are in float values

  • Categories 14 to 17 are higher



sns.countplot(df['City_Category'])
  • Higher count might represent the urban area indicates more population



sns.countplot(df['Stay_In_Current_City_Years'])
  • Most buyers have one year living in the city

  • Remaining categories are uniform distribution



Now let us plot using two variables for analysis

# bivariate analysis
occupation_plot = df.pivot_table(index='Occupation', values='Purchase', aggfunc=np.mean)
occupation_plot.plot(kind='bar', figsize=(13, 7))
plt.xlabel('Occupation')
plt.ylabel("Purchase")
plt.title("Occupation and Purchase Analysis")
plt.xticks(rotation=0)
plt.show()
  • np.mean will display mean of the purchase based on occupation

  • np.sum will display a sum of the purchase based on occupation

  • Based on the labels, we can observe all the categories being purchased in an average manner.

  • Recommended plot graph for presentation



age_plot = df.pivot_table(index='Age', values='Purchase', aggfunc=np.mean)
age_plot.plot(kind='bar', figsize=(13, 7))
plt.xlabel('Age')
plt.ylabel("Purchase")
plt.title("Age and Purchase Analysis")
plt.xticks(rotation=0)
plt.show()
  • Age and Purchase graph also has a uniform distribution.



gender_plot = df.pivot_table(index='Gender', values='Purchase', aggfunc=np.mean)
gender_plot.plot(kind='bar', figsize=(13, 7))
plt.xlabel('Gender')
plt.ylabel("Purchase")
plt.title("Gender and Purchase Analysis")
plt.xticks(rotation=0)
plt.show()
  • Uniform distribution but with a little difference



Preprocessing the dataset


We must check first for null values in the data

# check for null values
df.isnull().sum()
  • Null values are present in Product_Category_2 and Product_Category_3

  • Null values must be filled for easier processing



Now we fill the Null values in the dataset

df['Product_Category_2'] = df['Product_Category_2'].fillna(-2.0).astype("float32")
df['Product_Category_3'] = df['Product_Category_3'].fillna(-2.0).astype("float32")
  • Null values filled with a negative value to not affect the results.

  • The value filled must be of same data type of the attribute.


Let us double check the null values

df.isnull().sum()


Now we must convert the categorical attributes to numerical using a dictionary

# encoding values using dict
gender_dict = {'F':0, 'M':1}
df['Gender'] = df['Gender'].apply(lambda x: gender_dict[x])
df.head()
  • 'F' now converted to numerical zero (0), same for 'M' to one (1)


Label encoding is to convert the categorical column into the numerical column a lot quicker


# to improve the metric use one hot encoding
# label encoding
cols = ['Age', 'City_Category', 'Stay_In_Current_City_Years']
from sklearn.preprocessing import LabelEncoder
le = LabelEncoder()
for col in cols:
    df[col] = le.fit_transform(df[col])
df.head()
  • One hot encoding increases the no. of columns but improves accuracy

  • More columns means more data to train, it will increase the training time

  • All categorical columns converted to numerical

  • For the input User_ID and Product_ID must be removed in order to generalize the results.



Coorelation Matrix


A correlation matrix is a table showing correlation coefficients between variables. Each cell in the table shows the correlation between two variables. The value is in the range of -1 to 1. If two variables have a high correlation, we can neglect one variable from those two.


corr = df.corr()
plt.figure(figsize=(14,7))
sns.heatmap(corr, annot=True, cmap='coolwarm')
  • Purchase is most correlated to Product_Category_1 and Product_Category_3

  • Marital_Status and Age also has positive correlation



Input Split


df.head()
  • User_ID and Product_ID must be removed for better results, if not the results will be biased to User_ID or Product_ID


Now we split the data for training

X = df.drop(columns=['User_ID', 'Product_ID', 'Purchase'])
y = df['Purchase']
  • Purchase is an output data that is why it is removed from X as well



Model Training


from sklearn.model_selection import cross_val_score, train_test_split
from sklearn.metrics import mean_squared_error
def train(model, X, y):
    # train-test split
    x_train, x_test, y_train, y_test = train_test_split(X, y,         
random_state=42, test_size=0.25)
    model.fit(x_train, y_train)
    
    # predict the results
    pred = model.predict(x_test)
    
    # cross validation
    cv_score = cross_val_score(model, X, y, scoring='neg_mean_squared_error', cv=5)
    cv_score = np.abs(np.mean(cv_score))
    
    print("Results")
    print("MSE:", np.sqrt(mean_squared_error(y_test, pred)))
    print("CV Score:", np.sqrt(cv_score))
  • cross val score() is used for better validation of the model.

  • cv=5 means that the cross-validation will split the data into 5 parts for training.

  • np.abs() will convert the negative score to positive and np.mean() will give the average value of 5 scores.



Now we display the basic models

from sklearn.linear_model import LinearRegression
model = LinearRegression(normalize=True)
train(model, X, y)
coef = pd.Series(model.coef_, X.columns).sort_values()
coef.plot(kind='bar', title='Model Coefficients')

Results MSE: 4617.994034201719 CV Score: 4625.252945835687

  • Linear Regression model must have normalized data to give better results

  • Gender category has high coefficient for the Linear Regression model



from sklearn.tree import DecisionTreeRegressor
model = DecisionTreeRegressor()
train(model, X, y)
features = pd.Series(model.feature_importances_, X.columns).sort_values(ascending=False)
features.plot(kind='bar', title='Feature Importance')

Results MSE: 3366.9672356860747 CV Score: 3338.5905886644855

  • Results have improved compared to Linear Regression model

  • Product_Category_1 has high feature importance compared to the Linear Regression model.



from sklearn.ensemble import RandomForestRegressor
model = RandomForestRegressor(n_jobs=-1)
train(model, X, y)
features = pd.Series(model.feature_importances_, X.columns).sort_values(ascending=False)
features.plot(kind='bar', title='Feature Importance')

Results MSE: 3062.66041010778 CV Score: 3052.7778119222253

  • Better results compared with Decision Tree Regressor



from sklearn.ensemble import ExtraTreesRegressor
model = ExtraTreesRegressor(n_jobs=-1)
train(model, X, y)
features = pd.Series(model.feature_importances_, X.columns).sort_values(ascending=False)
features.plot(kind='bar', title='Feature Importance&