import pandas as pd import datetime from scipy import stats import matplotlib.pyplot as plt from matplotlib import cm import plotly.graph_objects as go import seaborn as sns import plotly.express as px from matplotlib import colors from matplotlib.colors import ListedColormap import numpy as np from numpy import nan from sklearn.preprocessing import StandardScaler from sklearn.decomposition import PCA from yellowbrick.cluster import KElbowVisualizer from sklearn.cluster import KMeans from sklearn.metrics import silhouette_score

market_df = pd.read_csv('marketing_campaign.csv', sep='\t', lineterminator='\r')

market_df['Dt_Customer'] = pd.to_datetime(market_df['Dt_Customer'], format="%d-%m-%Y")

market_df

market_df.info()

market_df.describe(include='all')

market_df.isnull().any()

filtered_df = market_df.copy() filtered_df.head(20)

# Dropping redundant columns filtered_df.drop(['ID', 'Dt_Customer', 'Z_CostContact', 'Z_Revenue'], axis=1, inplace=True) filtered_df.drop(filtered_df.tail(1).index,inplace=True) # Last row was full of NaN values filtered_df

# Filling NaN values with median considering Education and Marital_Status filtered_df['Income'] = filtered_df['Income'].fillna(filtered_df.groupby(['Education', 'Marital_Status'])['Income'].transform('mean'))

filtered_df.isnull().sum() # We don't have any NaN values. Time to crunch the numbers

# Convert year of birth to Age filtered_df = filtered_df.rename(columns={'Year_Birth': 'Age'}) filtered_df['Age'] = 2022 - filtered_df['Age'] # Adding amount spent filtered_df['Amount_Spent'] = filtered_df['MntWines'] + filtered_df['MntFruits'] + filtered_df['MntMeatProducts'] + filtered_df['MntFishProducts'] + filtered_df['MntSweetProducts'] + filtered_df['MntGoldProds'] filtered_df.drop(['MntWines', 'MntFruits', 'MntMeatProducts', 'MntFishProducts', 'MntSweetProducts', 'MntGoldProds'], axis=1, inplace=True) # Adding number of childs filtered_df['Children'] = filtered_df['Kidhome'] + filtered_df['Teenhome'] filtered_df.drop(['Kidhome', 'Teenhome'], axis=1, inplace=True) # Adding number of purchases filtered_df['Purchases'] = filtered_df['NumWebPurchases'] + filtered_df['NumCatalogPurchases'] + filtered_df['NumStorePurchases'] filtered_df.drop(['NumWebPurchases', 'NumCatalogPurchases', 'NumStorePurchases'], axis=1, inplace=True) filtered_df

# Grouping and dropping rows filled with absurd values filtered_df['Marital_Status'] = filtered_df['Marital_Status'].replace({'Alone': 'Single', 'Widow': 'Single', 'YOLO': 'Single', 'Divorced': 'Single', 'Married': 'Together'}) filtered_df = filtered_df.drop(filtered_df.loc[filtered_df['Marital_Status']=='Absurd'].index) filtered_df['Education']=filtered_df['Education'].replace({'Basic': 'Undergraduate', '2n Cycle': 'Undergraduate', 'Graduation': 'Graduate', 'Master': 'Postgraduate', 'PhD': 'Postgraduate'}) # Using quantiles to delete outliers income_upper_limit = filtered_df['Income'].quantile(0.99) income_lower_limit = filtered_df['Income'].quantile(0.01) filtered_df = filtered_df[(filtered_df['Income'] <= income_upper_limit) & (filtered_df['Income'] >= income_lower_limit)] age_lower_limit = filtered_df['Age'].quantile(0.99) filtered_df = filtered_df[(filtered_df['Age'] <= age_lower_limit)] # Drop some rows which were duplicates filtered_df = filtered_df.drop_duplicates(subset=['Income','Education','Marital_Status','Age'], ignore_index=True) filtered_df

# Checking correlation between features correlation_mat = filtered_df.corr() plt.figure(figsize=(20,20)) cmap = colors.ListedColormap(["#682F2F", "#9E726F", "#D6B2B1", "#B9C0C9", "#9F8A78", "#F3AB60"]) sns.heatmap(correlation_mat,annot=True, cmap=cmap, center=0)

# One hot encoding using get_dummies on the object type features object_cols = filtered_df.select_dtypes(['object']).columns X = pd.get_dummies(filtered_df, columns=object_cols) # Selecting all numerical features numeric_cols = filtered_df.select_dtypes(['float']).columns # Scaling data standard_scaler = StandardScaler() standard_scaler.fit(X[numeric_cols]) X[numeric_cols] = standard_scaler.transform(X[numeric_cols])

X

# PCA is used for Dimensionality Reduction and removes Multicollinearity between features pca = PCA(n_components=3) pca.fit(X) PCA_X = pd.DataFrame(pca.transform(X), columns=['col1', 'col2', 'col3']) # Elbow technique to choose the right number of clusters elbow = KElbowVisualizer(KMeans(), k=10) elbow.fit(PCA_X) elbow.show()

# KMeans clustering kmeans = KMeans(n_clusters=4, random_state=42) predict = kmeans.fit_predict(PCA_X) filtered_df['Clusters'] = predict

labels = ['Cluster 0', 'Cluster 1', 'Cluster 2', 'Cluster 3'] cluster0_val = filtered_df[filtered_df["Clusters"]==0].shape[0] cluster1_val = filtered_df[filtered_df["Clusters"]==1].shape[0] cluster2_val = filtered_df[filtered_df["Clusters"]==2].shape[0] cluster3_val = filtered_df[filtered_df["Clusters"]==3].shape[0] values = [cluster0_val, cluster1_val, cluster2_val, cluster3_val] fig = go.Figure(data=[go.Pie(labels=labels, values=values, hole = 0.5, title="Clusters")]) fig.show()

plot = sns.scatterplot(x = filtered_df['Amount_Spent'], y = filtered_df['Income'], hue = filtered_df['Clusters']) plot.set_title('Clusters based on income and amount spent') plt.legend() plt.show()

plt.figure(figsize=(10, 6)) plot = sns.boxplot(x = filtered_df['Clusters'], y = filtered_df['Purchases'], hue = filtered_df['Clusters']) plot.set_title('Clusters purchases') plt.legend() plt.show()

ax = plt.subplot() ax.bar(filtered_df['Clusters'], filtered_df['Purchases'], width=0.2, color='b', align='center', tick_label=filtered_df['Clusters'], label = 'Purchases') ax.bar(filtered_df['Clusters'], filtered_df['NumDealsPurchases'], width=0.2, color='g', align='center', tick_label=filtered_df['Clusters'], label = 'Deals') ax.set_title('Purchases during deals') plt.xlabel('Cluster') plt.ylabel('Deals / Purchases') ax.legend() plt.show()

plt.figure(figsize=(10, 6)) plot = sns.boxplot(x = filtered_df['Clusters'], y = filtered_df['Age'], hue = filtered_df['Clusters']) plot.set_title('Clusters age repartition') plt.legend() plt.show()

plt.figure(figsize=(10, 6)) plot = sns.barplot(x = filtered_df['Clusters'], y = filtered_df['Children'], hue = filtered_df['Clusters']) plot.set_title('Clusters childs number') plt.legend() plt.show()