Introduction¶
Welcome to the first installment of my role-play series as a data expert in the iGaming industry. All individuals and organizations mentioned in this series are fictional and created solely for this exercise. To avoid misinformation, I will not share this outside of this context.
Churn Analysis and Predictive Modeling in iGaming¶
For my first article, I chose churn analysis and predictive modeling, a favorite topic in industries with high customer turnover. In any relationship, predicting the other party’s feelings is crucial. In this series, I worked with a comprehensive iGaming dataset to predict whether a customer: is about to leave, or has already disconnected from our services.
Objective¶
The goal is to build a predictive model that can:
- Analyze user data.
- Determine the relationship status:
- ✅ Intact
- ⚠️ Weakening
- ❌ Broken
- Assess customer status in real-time or forecast future churn risks.
Dataset Overview¶
The dataset includes the following key features:
Core Identifiers¶
PlayerIDChurn(Target:Yes/No)
Financial Activity¶
MonthlyDepositsMonthlyWageringTotalBonusTakenOverageWagering
Engagement & Support¶
VIPConciergeContactsCustomerCareCallsRetentionCallContacts
Behavioral Metrics¶
CrossBorderPlayPeakTimePlayOffPeakPlayActiveDays
Technical & Transactional¶
FailedTransactionsPendingPayoutsGameSessionRedirects
Demographics & Preferences¶
AgeOfCustomerPlayerTierGeoSegmentMaritalStatus
Risk & Loyalty Indicators¶
HighRiskBetsSOI(Statement of Intent)SOW(Scope of Work)OptOutPromotions
import library
In [241]:
import numpy as np
import pylab as pl
import pandas as pd
import matplotlib.pyplot as plt
%matplotlib inline
import seaborn as sns
from sklearn.utils import shuffle
from sklearn.svm import SVC
from sklearn.metrics import confusion_matrix,classification_report
from sklearn.model_selection import cross_val_score, GridSearchCV
# Input data files are available in the "../input/" directory.
# For example, running this (by clicking run or pressing Shift+Enter) will list the files in the input directory
# Any results you write to the current directory are saved as output.
read dataset
In [242]:
train = pd.read_csv("C:/Users/Memre/VIPuserdatatrain.csv")
test = pd.read_csv("C:/Users/Memre/VIPuserdata.csv")
In [243]:
train.info()
train[0:10]
<class 'pandas.core.frame.DataFrame'> RangeIndex: 51047 entries, 0 to 51046 Data columns (total 58 columns): # Column Non-Null Count Dtype --- ------ -------------- ----- 0 PlayerID 51047 non-null int64 1 Churn 51047 non-null object 2 MonthlyDeposits 50891 non-null float64 3 MonthlyWagering 50891 non-null float64 4 TotalBonusTaken 50891 non-null float64 5 VIPConciergeContacts 50891 non-null float64 6 OverageWagering 50891 non-null float64 7 CrossBorderPlay 50891 non-null float64 8 PercChangeWagering 50680 non-null float64 9 PercChangeDeposits 50680 non-null float64 10 DroppedCalls 51047 non-null float64 11 BlockedCalls 51047 non-null float64 12 UnansweredCalls 51047 non-null float64 13 CustomerCareCalls 51047 non-null float64 14 ThreewayCalls 51047 non-null float64 15 ReceivedCalls 51047 non-null float64 16 OutboundCalls 51047 non-null float64 17 InboundCalls 51047 non-null float64 18 PeakTimePlay 51047 non-null float64 19 OffPeakPlay 51047 non-null float64 20 FailedTransactions 51047 non-null float64 21 GameSessionRedirects 51047 non-null float64 22 PendingPayouts 51047 non-null float64 23 MonthsActive 51047 non-null int64 24 UniqueGamesPlayed 51047 non-null int64 25 ActiveDays 51047 non-null int64 26 GameCategoryPref 51023 non-null object 27 DevicesUsed 51046 non-null float64 28 GameTitlesPlayed 51046 non-null float64 29 DaysSinceLastDeviceChange 51046 non-null float64 30 AgePrimaryAccount 50138 non-null float64 31 AgeSecondaryUser 50138 non-null float64 32 HouseholdSize 51047 non-null object 33 UsedPromoCode 51047 non-null object 34 MobileCapable 51047 non-null object 35 HighRiskBets 51047 non-null object 36 SOI 51047 non-null object 37 SOW 51047 non-null object 38 OfflineDepositUser 51047 non-null object 39 RespondsToVIPOffers 51047 non-null object 40 OptOutPromotions 51047 non-null object 41 IntlPlayFlag 51047 non-null object 42 DesktopUser 51047 non-null object 43 HasActivePaymentMethod 51047 non-null object 44 RetentionCallContacts 51047 non-null int64 45 RetentionOffersAccepted 51047 non-null int64 46 NewPlayerFlag 51047 non-null object 47 ExistingPlayerFlag 51047 non-null object 48 ReferralsGenerated 51047 non-null int64 49 PlayerTier 51047 non-null int64 50 HighRiskCasino 51047 non-null object 51 CreditAdjustments 51047 non-null int64 52 AvgBetSize 51047 non-null object 53 ContactedRetention 51047 non-null object 54 UserPotentialTier 51047 non-null object 55 GeoSegment 51047 non-null object 56 EmploymentStatus 51047 non-null object 57 MaritalStatus 51047 non-null object dtypes: float64(26), int64(9), object(23) memory usage: 22.6+ MB
Out[243]:
| PlayerID | Churn | MonthlyDeposits | MonthlyWagering | TotalBonusTaken | VIPConciergeContacts | OverageWagering | CrossBorderPlay | PercChangeWagering | PercChangeDeposits | ... | ReferralsGenerated | PlayerTier | HighRiskCasino | CreditAdjustments | AvgBetSize | ContactedRetention | UserPotentialTier | GeoSegment | EmploymentStatus | MaritalStatus | |
|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|
| 0 | 3000002 | Yes | 24.00 | 219.0 | 22.0 | 0.25 | 0.0 | 0.0 | -157.0 | -19.0 | ... | 0 | 4 | No | 0 | 30 | Yes | 1-Highest | Suburban | Professional | No |
| 1 | 3000010 | Yes | 16.99 | 10.0 | 17.0 | 0.00 | 0.0 | 0.0 | -4.0 | 0.0 | ... | 0 | 5 | No | 0 | 30 | No | 4-Medium | Suburban | Professional | Yes |
| 2 | 3000014 | No | 38.00 | 8.0 | 38.0 | 0.00 | 0.0 | 0.0 | -2.0 | 0.0 | ... | 0 | 6 | No | 0 | Unknown | No | 3-Good | Town | Crafts | Yes |
| 3 | 3000022 | No | 82.28 | 1312.0 | 75.0 | 1.24 | 0.0 | 0.0 | 157.0 | 8.1 | ... | 0 | 6 | No | 0 | 10 | No | 4-Medium | Other | Other | No |
| 4 | 3000026 | Yes | 17.14 | 0.0 | 17.0 | 0.00 | 0.0 | 0.0 | 0.0 | -0.2 | ... | 0 | 9 | No | 1 | 10 | No | 1-Highest | Other | Professional | Yes |
| 5 | 3000030 | No | 38.05 | 682.0 | 52.0 | 0.25 | 0.0 | 0.0 | 148.0 | -3.1 | ... | 0 | 1 | No | 1 | 30 | No | 3-Good | Other | Other | Yes |
| 6 | 3000038 | No | 31.66 | 26.0 | 30.0 | 0.25 | 0.0 | 0.0 | 60.0 | 4.0 | ... | 0 | 9 | No | 1 | 30 | No | 1-Highest | Other | Self | Yes |
| 7 | 3000042 | No | 62.13 | 98.0 | 66.0 | 2.48 | 0.0 | 0.0 | 24.0 | 6.8 | ... | 0 | 6 | No | 0 | 30 | No | 1-Highest | Other | Professional | No |
| 8 | 3000046 | No | 35.30 | 24.0 | 35.0 | 0.00 | 0.0 | 0.0 | 20.0 | -0.3 | ... | 0 | 9 | No | 0 | 80 | No | 1-Highest | Other | Other | Yes |
| 9 | 3000050 | No | 81.00 | 1056.0 | 75.0 | 0.00 | 0.0 | 0.0 | 43.0 | 2.4 | ... | 0 | 5 | No | 1 | 30 | No | 3-Good | Other | Professional | No |
10 rows × 58 columns
In [244]:
#Churn : Yes:1 , No:0
Churn = {'Yes': 1,'No': 0}
# traversing through dataframe
# values where key matches
train.Churn = [Churn[item] for item in train.Churn]
print(train)
PlayerID Churn MonthlyDeposits MonthlyWagering TotalBonusTaken \
0 3000002 1 24.00 219.0 22.0
1 3000010 1 16.99 10.0 17.0
2 3000014 0 38.00 8.0 38.0
3 3000022 0 82.28 1312.0 75.0
4 3000026 1 17.14 0.0 17.0
... ... ... ... ... ...
51042 3399958 1 NaN NaN NaN
51043 3399974 0 95.17 1745.0 85.0
51044 3399978 1 NaN NaN NaN
51045 3399990 0 NaN NaN NaN
51046 3399994 0 NaN NaN NaN
VIPConciergeContacts OverageWagering CrossBorderPlay \
0 0.25 0.0 0.0
1 0.00 0.0 0.0
2 0.00 0.0 0.0
3 1.24 0.0 0.0
4 0.00 0.0 0.0
... ... ... ...
51042 NaN NaN NaN
51043 0.99 45.0 4.7
51044 NaN NaN NaN
51045 NaN NaN NaN
51046 NaN NaN NaN
PercChangeWagering PercChangeDeposits ... ReferralsGenerated \
0 -157.0 -19.0 ... 0
1 -4.0 0.0 ... 0
2 -2.0 0.0 ... 0
3 157.0 8.1 ... 0
4 0.0 -0.2 ... 0
... ... ... ... ...
51042 NaN NaN ... 0
51043 122.0 15.9 ... 0
51044 NaN NaN ... 0
51045 NaN NaN ... 0
51046 NaN NaN ... 0
PlayerTier HighRiskCasino CreditAdjustments AvgBetSize \
0 4 No 0 30
1 5 No 0 30
2 6 No 0 Unknown
3 6 No 0 10
4 9 No 1 10
... ... ... ... ...
51042 6 No 0 60
51043 9 No 1 60
51044 7 No 1 80
51045 9 No 0 30
51046 0 No 1 60
ContactedRetention UserPotentialTier GeoSegment EmploymentStatus \
0 Yes 1-Highest Suburban Professional
1 No 4-Medium Suburban Professional
2 No 3-Good Town Crafts
3 No 4-Medium Other Other
4 No 1-Highest Other Professional
... ... ... ... ...
51042 No 1-Highest Suburban Other
51043 No 3-Good Other Other
51044 No 5-Low Other Clerical
51045 No 5-Low Other Other
51046 Yes 5-Low Other Other
MaritalStatus
0 No
1 Yes
2 Yes
3 No
4 Yes
... ...
51042 Yes
51043 No
51044 No
51045 No
51046 Unknown
[51047 rows x 58 columns]
Handling missing data¶
In [245]:
print("Any missing sample in training set:",train.isnull().values.any())
print("Any missing sample in test set:",test.isnull().values.any(), "\n")
Any missing sample in training set: True Any missing sample in test set: True
We replaced all missing values with zeros for simplicity in this POC. While other methods like imputation exist, this approach keeps things straightforward.
In [246]:
# for column
#train['MonthlyRevenue'].fillna((train['MonthlyRevenue'].median()), inplace=True)
# for column
train['MonthlyDeposits'] = train['MonthlyDeposits'].replace(np.nan, 0)
# for whole dataframe
train = train.replace(np.nan, 0)
# inplace
train.replace(np.nan, 0, inplace=True)
In [247]:
# for column
#train['MonthlyMinutes'].fillna((train['MonthlyMinutes'].median()), inplace=True)
train['MonthlyWagering'] = train['MonthlyWagering'].replace(np.nan, 0)
# for whole dataframe
train = train.replace(np.nan, 0)
# inplace
train.replace(np.nan, 0, inplace=True)
In [248]:
# for column
#train['TotalRecurringCharge'].fillna((train['TotalRecurringCharge'].median()), inplace=True)
train['TotalBonusTaken'] = train['TotalBonusTaken'].replace(np.nan, 0)
# for whole dataframe
train = train.replace(np.nan, 0)
# inplace
train.replace(np.nan, 0, inplace=True)
In [249]:
# for column
#train['DirectorAssistedCalls'].fillna((train['DirectorAssistedCalls'].median()), inplace=True)
train['VIPConciergeContacts'] = train['VIPConciergeContacts'].replace(np.nan, 0)
# for whole dataframe
train = train.replace(np.nan, 0)
# inplace
train.replace(np.nan, 0, inplace=True)
print(train)
PlayerID Churn MonthlyDeposits MonthlyWagering TotalBonusTaken \
0 3000002 1 24.00 219.0 22.0
1 3000010 1 16.99 10.0 17.0
2 3000014 0 38.00 8.0 38.0
3 3000022 0 82.28 1312.0 75.0
4 3000026 1 17.14 0.0 17.0
... ... ... ... ... ...
51042 3399958 1 0.00 0.0 0.0
51043 3399974 0 95.17 1745.0 85.0
51044 3399978 1 0.00 0.0 0.0
51045 3399990 0 0.00 0.0 0.0
51046 3399994 0 0.00 0.0 0.0
VIPConciergeContacts OverageWagering CrossBorderPlay \
0 0.25 0.0 0.0
1 0.00 0.0 0.0
2 0.00 0.0 0.0
3 1.24 0.0 0.0
4 0.00 0.0 0.0
... ... ... ...
51042 0.00 0.0 0.0
51043 0.99 45.0 4.7
51044 0.00 0.0 0.0
51045 0.00 0.0 0.0
51046 0.00 0.0 0.0
PercChangeWagering PercChangeDeposits ... ReferralsGenerated \
0 -157.0 -19.0 ... 0
1 -4.0 0.0 ... 0
2 -2.0 0.0 ... 0
3 157.0 8.1 ... 0
4 0.0 -0.2 ... 0
... ... ... ... ...
51042 0.0 0.0 ... 0
51043 122.0 15.9 ... 0
51044 0.0 0.0 ... 0
51045 0.0 0.0 ... 0
51046 0.0 0.0 ... 0
PlayerTier HighRiskCasino CreditAdjustments AvgBetSize \
0 4 No 0 30
1 5 No 0 30
2 6 No 0 Unknown
3 6 No 0 10
4 9 No 1 10
... ... ... ... ...
51042 6 No 0 60
51043 9 No 1 60
51044 7 No 1 80
51045 9 No 0 30
51046 0 No 1 60
ContactedRetention UserPotentialTier GeoSegment EmploymentStatus \
0 Yes 1-Highest Suburban Professional
1 No 4-Medium Suburban Professional
2 No 3-Good Town Crafts
3 No 4-Medium Other Other
4 No 1-Highest Other Professional
... ... ... ... ...
51042 No 1-Highest Suburban Other
51043 No 3-Good Other Other
51044 No 5-Low Other Clerical
51045 No 5-Low Other Other
51046 Yes 5-Low Other Other
MaritalStatus
0 No
1 Yes
2 Yes
3 No
4 Yes
... ...
51042 Yes
51043 No
51044 No
51045 No
51046 Unknown
[51047 rows x 58 columns]
In [250]:
from sklearn.preprocessing import LabelEncoder
le = LabelEncoder()
def FunLabelEncoder(df):
for c in df.columns:
if df.dtypes[c] == object:
le.fit(df[c].astype(str))
df[c] = le.transform(df[c].astype(str))
return df
In [251]:
train = FunLabelEncoder(train)
train.info()
train.iloc[235:300,:]
<class 'pandas.core.frame.DataFrame'> RangeIndex: 51047 entries, 0 to 51046 Data columns (total 58 columns): # Column Non-Null Count Dtype --- ------ -------------- ----- 0 PlayerID 51047 non-null int64 1 Churn 51047 non-null int64 2 MonthlyDeposits 51047 non-null float64 3 MonthlyWagering 51047 non-null float64 4 TotalBonusTaken 51047 non-null float64 5 VIPConciergeContacts 51047 non-null float64 6 OverageWagering 51047 non-null float64 7 CrossBorderPlay 51047 non-null float64 8 PercChangeWagering 51047 non-null float64 9 PercChangeDeposits 51047 non-null float64 10 DroppedCalls 51047 non-null float64 11 BlockedCalls 51047 non-null float64 12 UnansweredCalls 51047 non-null float64 13 CustomerCareCalls 51047 non-null float64 14 ThreewayCalls 51047 non-null float64 15 ReceivedCalls 51047 non-null float64 16 OutboundCalls 51047 non-null float64 17 InboundCalls 51047 non-null float64 18 PeakTimePlay 51047 non-null float64 19 OffPeakPlay 51047 non-null float64 20 FailedTransactions 51047 non-null float64 21 GameSessionRedirects 51047 non-null float64 22 PendingPayouts 51047 non-null float64 23 MonthsActive 51047 non-null int64 24 UniqueGamesPlayed 51047 non-null int64 25 ActiveDays 51047 non-null int64 26 GameCategoryPref 51047 non-null int64 27 DevicesUsed 51047 non-null float64 28 GameTitlesPlayed 51047 non-null float64 29 DaysSinceLastDeviceChange 51047 non-null float64 30 AgePrimaryAccount 51047 non-null float64 31 AgeSecondaryUser 51047 non-null float64 32 HouseholdSize 51047 non-null int64 33 UsedPromoCode 51047 non-null int64 34 MobileCapable 51047 non-null int64 35 HighRiskBets 51047 non-null int64 36 SOI 51047 non-null int64 37 SOW 51047 non-null int64 38 OfflineDepositUser 51047 non-null int64 39 RespondsToVIPOffers 51047 non-null int64 40 OptOutPromotions 51047 non-null int64 41 IntlPlayFlag 51047 non-null int64 42 DesktopUser 51047 non-null int64 43 HasActivePaymentMethod 51047 non-null int64 44 RetentionCallContacts 51047 non-null int64 45 RetentionOffersAccepted 51047 non-null int64 46 NewPlayerFlag 51047 non-null int64 47 ExistingPlayerFlag 51047 non-null int64 48 ReferralsGenerated 51047 non-null int64 49 PlayerTier 51047 non-null int64 50 HighRiskCasino 51047 non-null int64 51 CreditAdjustments 51047 non-null int64 52 AvgBetSize 51047 non-null int64 53 ContactedRetention 51047 non-null int64 54 UserPotentialTier 51047 non-null int64 55 GeoSegment 51047 non-null int64 56 EmploymentStatus 51047 non-null int64 57 MaritalStatus 51047 non-null int64 dtypes: float64(26), int64(32) memory usage: 22.6 MB
Out[251]:
| PlayerID | Churn | MonthlyDeposits | MonthlyWagering | TotalBonusTaken | VIPConciergeContacts | OverageWagering | CrossBorderPlay | PercChangeWagering | PercChangeDeposits | ... | ReferralsGenerated | PlayerTier | HighRiskCasino | CreditAdjustments | AvgBetSize | ContactedRetention | UserPotentialTier | GeoSegment | EmploymentStatus | MaritalStatus | |
|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|
| 235 | 3001782 | 1 | 64.80 | 479.0 | 50.0 | 0.00 | 139.0 | 0.0 | -337.0 | -13.3 | ... | 0 | 3 | 0 | 0 | 8 | 0 | 0 | 2 | 4 | 2 |
| 236 | 3001786 | 1 | 17.14 | 3.0 | 17.0 | 0.00 | 0.0 | 0.0 | 0.0 | -0.2 | ... | 0 | 9 | 0 | 0 | 15 | 0 | 0 | 2 | 3 | 2 |
| 237 | 3001790 | 0 | 124.06 | 2955.0 | 82.0 | 0.00 | 421.0 | 0.0 | 262.0 | 20.0 | ... | 0 | 0 | 0 | 1 | 14 | 0 | 0 | 2 | 3 | 1 |
| 238 | 3001794 | 0 | 80.28 | 1245.0 | 68.0 | 1.24 | 73.0 | 0.0 | -160.0 | -62.1 | ... | 0 | 6 | 0 | 3 | 0 | 0 | 2 | 0 | 4 | 2 |
| 239 | 3001802 | 0 | 100.05 | 820.0 | 85.0 | 3.46 | 0.0 | 17.5 | 2.0 | -16.1 | ... | 0 | 8 | 0 | 0 | 0 | 0 | 0 | 0 | 4 | 2 |
| ... | ... | ... | ... | ... | ... | ... | ... | ... | ... | ... | ... | ... | ... | ... | ... | ... | ... | ... | ... | ... | ... |
| 295 | 3002250 | 1 | 95.30 | 894.0 | 52.0 | 2.48 | 30.0 | 28.4 | 146.0 | -10.0 | ... | 0 | 0 | 0 | 0 | 3 | 0 | 2 | 0 | 3 | 1 |
| 296 | 3002258 | 0 | 104.29 | 1005.0 | 105.0 | 0.99 | 2.0 | 3.7 | -145.0 | 1.9 | ... | 0 | 0 | 0 | 0 | 8 | 0 | 0 | 0 | 3 | 1 |
| 297 | 3002262 | 0 | 95.27 | 815.0 | 95.0 | 1.49 | 0.0 | 0.0 | -350.0 | -7.0 | ... | 0 | 5 | 0 | 0 | 0 | 0 | 0 | 0 | 3 | 1 |
| 298 | 3002270 | 0 | 79.84 | 596.0 | 75.0 | 1.73 | 8.0 | 0.0 | -38.0 | -0.9 | ... | 0 | 9 | 0 | 0 | 3 | 0 | 0 | 2 | 3 | 2 |
| 299 | 3002274 | 1 | 55.34 | 144.0 | 55.0 | 1.73 | 0.0 | 0.0 | -137.0 | -55.3 | ... | 0 | 9 | 0 | 0 | 14 | 0 | 0 | 0 | 3 | 0 |
65 rows × 58 columns
In [252]:
test = FunLabelEncoder(test)
test.info()
test.iloc[235:300,:]
<class 'pandas.core.frame.DataFrame'> RangeIndex: 20000 entries, 0 to 19999 Data columns (total 58 columns): # Column Non-Null Count Dtype --- ------ -------------- ----- 0 PlayerID 20000 non-null int64 1 Churn 0 non-null float64 2 MonthlyDeposits 19940 non-null float64 3 MonthlyWagering 19940 non-null float64 4 TotalBonusTaken 19940 non-null float64 5 VIPConciergeContacts 19940 non-null float64 6 OverageWagering 19940 non-null float64 7 CrossBorderPlay 19940 non-null float64 8 PercChangeWagering 19865 non-null float64 9 PercChangeDeposits 19865 non-null float64 10 DroppedCalls 20000 non-null float64 11 BlockedCalls 20000 non-null float64 12 UnansweredCalls 20000 non-null float64 13 CustomerCareCalls 20000 non-null float64 14 ThreewayCalls 20000 non-null float64 15 ReceivedCalls 20000 non-null float64 16 OutboundCalls 20000 non-null float64 17 InboundCalls 20000 non-null float64 18 PeakTimePlay 20000 non-null float64 19 OffPeakPlay 20000 non-null float64 20 FailedTransactions 20000 non-null float64 21 GameSessionRedirects 20000 non-null float64 22 PendingPayouts 20000 non-null float64 23 MonthsActive 20000 non-null int64 24 UniqueGamesPlayed 20000 non-null int64 25 ActiveDays 20000 non-null int64 26 GameCategoryPref 20000 non-null int64 27 DevicesUsed 20000 non-null int64 28 GameTitlesPlayed 20000 non-null int64 29 DaysSinceLastDeviceChange 20000 non-null int64 30 AgePrimaryAccount 19665 non-null float64 31 AgeSecondaryUser 19665 non-null float64 32 HouseholdSize 20000 non-null int64 33 UsedPromoCode 20000 non-null int64 34 MobileCapable 20000 non-null int64 35 HighRiskBets 20000 non-null int64 36 SOI 20000 non-null int64 37 SOW 20000 non-null int64 38 OfflineDepositUser 20000 non-null int64 39 RespondsToVIPOffers 20000 non-null int64 40 OptOutPromotions 20000 non-null int64 41 IntlPlayFlag 20000 non-null int64 42 DesktopUser 20000 non-null int64 43 HasActivePaymentMethod 20000 non-null int64 44 RetentionCallContacts 20000 non-null int64 45 RetentionOffersAccepted 20000 non-null int64 46 NewPlayerFlag 20000 non-null int64 47 ExistingPlayerFlag 20000 non-null int64 48 ReferralsGenerated 20000 non-null int64 49 PlayerTier 20000 non-null int64 50 HighRiskCasino 20000 non-null int64 51 CreditAdjustments 20000 non-null int64 52 AvgBetSize 20000 non-null int64 53 ContactedRetention 20000 non-null int64 54 UserPotentialTier 20000 non-null int64 55 GeoSegment 20000 non-null int64 56 EmploymentStatus 20000 non-null int64 57 MaritalStatus 20000 non-null int64 dtypes: float64(24), int64(34) memory usage: 8.9 MB
Out[252]:
| PlayerID | Churn | MonthlyDeposits | MonthlyWagering | TotalBonusTaken | VIPConciergeContacts | OverageWagering | CrossBorderPlay | PercChangeWagering | PercChangeDeposits | ... | ReferralsGenerated | PlayerTier | HighRiskCasino | CreditAdjustments | AvgBetSize | ContactedRetention | UserPotentialTier | GeoSegment | EmploymentStatus | MaritalStatus | |
|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|
| 235 | 3004718 | NaN | 33.10 | 242.0 | 40.0 | 0.00 | 0.0 | 0.0 | -110.0 | -2.7 | ... | 0 | 4 | 0 | 0 | 8 | 0 | 0 | 2 | 1 | 2 |
| 236 | 3004722 | NaN | 64.69 | 370.0 | 40.0 | 0.25 | 98.0 | 0.0 | 8.0 | 0.1 | ... | 1 | 0 | 0 | 1 | 8 | 0 | 0 | 3 | 3 | 1 |
| 237 | 3004738 | NaN | 100.56 | 660.0 | 50.0 | 3.46 | 151.0 | 0.0 | -390.0 | -47.6 | ... | 0 | 6 | 0 | 0 | 3 | 0 | 2 | 0 | 3 | 0 |
| 238 | 3004746 | NaN | 35.17 | 308.0 | 30.0 | 0.00 | 14.0 | 0.0 | 42.0 | 10.9 | ... | 1 | 0 | 0 | 1 | 15 | 0 | 0 | 2 | 3 | 2 |
| 239 | 3004762 | NaN | 30.25 | 268.0 | 30.0 | 0.25 | 0.0 | 0.0 | 48.0 | -0.2 | ... | 0 | 6 | 0 | 0 | 3 | 0 | 0 | 2 | 4 | 2 |
| ... | ... | ... | ... | ... | ... | ... | ... | ... | ... | ... | ... | ... | ... | ... | ... | ... | ... | ... | ... | ... | ... |
| 295 | 3005682 | NaN | 81.49 | 741.0 | 75.0 | 5.69 | 0.0 | 0.8 | 185.0 | -1.5 | ... | 0 | 0 | 0 | 0 | 14 | 0 | 0 | 2 | 3 | 1 |
| 296 | 3005710 | NaN | 30.00 | 204.0 | 30.0 | 0.00 | 0.0 | 0.0 | 43.0 | 0.0 | ... | 0 | 0 | 0 | 0 | 8 | 0 | 0 | 0 | 3 | 1 |
| 297 | 3005750 | NaN | 51.21 | 312.0 | 35.0 | 0.25 | 37.0 | 0.0 | -149.0 | -13.0 | ... | 0 | 6 | 0 | 0 | 2 | 0 | 2 | 0 | 3 | 1 |
| 298 | 3005766 | NaN | 35.72 | 94.0 | 30.0 | 3.46 | 0.0 | 0.0 | -40.0 | -2.8 | ... | 0 | 7 | 0 | 1 | 8 | 0 | 0 | 2 | 3 | 0 |
| 299 | 3005834 | NaN | 38.29 | 252.0 | 45.0 | 1.73 | 0.0 | 0.0 | 88.0 | 2.7 | ... | 0 | 6 | 0 | 0 | 15 | 0 | 0 | 0 | 3 | 2 |
65 rows × 58 columns
In [253]:
test = test.drop(columns=['Churn'],
axis=1)
test = test.dropna(how='any')
print(test.shape)
(19533, 57)
In [254]:
#Frequency distribution of classes"
train_outcome = pd.crosstab(index=train["Churn"], # Make a crosstab
columns="count") # Name the count column
train_outcome
Out[254]:
| col_0 | count |
|---|---|
| Churn | |
| 0 | 36336 |
| 1 | 14711 |
In [255]:
# Distribution of Churn
train.Churn.value_counts()[0:30].plot(kind='bar')
plt.show()
Plotting Heatmap¶
A visual representation of data values using colors. Darker shades = stronger relationships. Helps spot key patterns at a glance.
In [256]:
train = train[["PlayerID","Churn","MonthlyDeposits","MonthlyWagering","TotalBonusTaken","VIPConciergeContacts","OverageWagering","CrossBorderPlay","PercChangeWagering","PercChangeDeposits","DroppedCalls","BlockedCalls","UnansweredCalls","CustomerCareCalls","ThreewayCalls","ReceivedCalls","OutboundCalls","InboundCalls","PeakTimePlay","OffPeakPlay","FailedTransactions","GameSessionRedirects","PendingPayouts","MonthsActive","UniqueGamesPlayed","ActiveDays","GameCategoryPref","DevicesUsed","GameTitlesPlayed","DaysSinceLastDeviceChange","AgePrimaryAccount","AgeSecondaryUser","HouseholdSize","UsedPromoCode","MobileCapable","HighRiskBets","SOI","SOW","OfflineDepositUser","RespondsToVIPOffers","OptOutPromotions","IntlPlayFlag","DesktopUser","HasActivePaymentMethod","RetentionCallContacts","RetentionOffersAccepted","NewPlayerFlag","ExistingPlayerFlag","ReferralsGenerated","PlayerTier","HighRiskCasino","CreditAdjustments","AvgBetSize","ContactedRetention","UserPotentialTier","GeoSegment","EmploymentStatus","MaritalStatus"]] #Subsetting the data
cor = train.corr() #Calculate the correlation of the above variables
sns.heatmap(cor, square = True) #Plot the correlation as heat map
Out[256]:
<Axes: >
Here appear important variables (customer churn behavior):
- TotalBonusTaken
- RoamingCalls
- DroppedCalls
- CustomerCareCalls
- OutboundCalls
- FailedTransactions
- PendingPayouts
- ActiveDays
- GameTitlesPlayed
- AgePrimaryAccount
- MobileCapable
- SOI
- OptOutPromotions
- HasActivePaymentMethod
- NewPlayerFlag
- PlayerTier
- AvgBetSize
- GeoSegment
SPLITING DATA¶
Data for training and testing To select a set of training data that will be input in the Machine Learning algorithm, to ensure that the classification algorithm training can be generalized well to new data. For this study using a sample size of 30%, assumed it ideal ratio between training and testing
In [257]:
from sklearn.model_selection import train_test_split
Y = train['Churn']
X = train.drop(columns=['Churn'])
X_train, X_test, Y_train, Y_test = train_test_split(X, Y, test_size=0.3, random_state=9)
In [258]:
print('X train shape: ', X_train.shape)
print('Y train shape: ', Y_train.shape)
print('X test shape: ', X_test.shape)
print('Y test shape: ', Y_test.shape)
X train shape: (35732, 57) Y train shape: (35732,) X test shape: (15315, 57) Y test shape: (15315,)
Random Forest for Churn Prediction¶
We're using Random Forest to predict which iGaming players might churn. Here's why it's effective:
How It Works¶
- Builds 100 decision trees, each analyzing players differently
- Each tree examines:
✓ Random subset of features (deposits, gameplay, etc.)
✓ Different sample of player data - Combines all trees' votes for more reliable predictions
Key Advantages¶
✔ Handles real-world data
- Robust to missing values
- Works with both numerical and categorical features
✔ Detects complex patterns
- Identifies interactions between factors (e.g., declining deposits + increased support calls)
✔ Built-in validation
- Uses out-of-bag samples to estimate accuracy during training
In [259]:
from sklearn.ensemble import RandomForestClassifier
# We define the model
rfcla = RandomForestClassifier(n_estimators=100,random_state=9,n_jobs=-1)
# We train model
rfcla.fit(X_train, Y_train)
# We predict target values
Y_predict5 = rfcla.predict(X_test)
In [260]:
# The confusion matrix
rfcla_cm = confusion_matrix(Y_test, Y_predict5)
f, ax = plt.subplots(figsize=(5,5))
sns.heatmap(rfcla_cm, annot=True, linewidth=0.7, linecolor='black', fmt='g', ax=ax, cmap="BuPu")
plt.title('Random Forest Classification Confusion Matrix')
plt.xlabel('Y predict')
plt.ylabel('Y test')
plt.show()
In [261]:
# Test score
score_rfcla = rfcla.score(X_test, Y_test)
print(score_rfcla)
0.7199477636304277
2. Naive Bayes Classification¶
Simple But Effective Probability¶
We're using Gaussian Naive Bayes to estimate churn probability based on player behavior patterns. It works by:
- Calculating how likely churn is given observed features (deposits, gameplay, etc.)
- Assuming features affect churn independently (the "naive" part)
- Modeling numerical data with a bell curve (Gaussian) distribution
Why We Considered It¶
✔ Fast training/prediction - Handles large datasets efficiently
✔ Works with incomplete data - Makes reasonable estimates with missing info
✔ Probabilistic outputs - Gives churn likelihood (0-100%) rather than just yes/no
In [262]:
from sklearn.naive_bayes import GaussianNB
# We define the model
nbcla = GaussianNB()
# We train model
nbcla.fit(X_train, Y_train)
# We predict target values
Y_predict3 = nbcla.predict(X_test)
In [263]:
# The confusion matrix
nbcla_cm = confusion_matrix(Y_test, Y_predict3)
f, ax = plt.subplots(figsize=(5,5))
sns.heatmap(nbcla_cm, annot=True, linewidth=0.7, linecolor='black', fmt='g', ax=ax, cmap="BuPu")
plt.title('Naive Bayes Classification Confusion Matrix')
plt.xlabel('Y predict')
plt.ylabel('Y test')
plt.show()
In [264]:
# Test score
score_nbcla = nbcla.score(X_test, Y_test)
print(score_nbcla)
0.6916748285994123
Comparison of classification techniques¶
Test score¶
In [265]:
Testscores = pd.Series([score_rfcla,score_nbcla, ],
index=['Random Forest Score','Naive Bayes Score' ])
print(Testscores)
Random Forest Score 0.719948 Naive Bayes Score 0.691675 dtype: float64
ROC Curve¶
is a graphical plot that illustrates the diagnostic ability of a binary classifier system as its discrimination threshold is varied.
In [266]:
from sklearn.metrics import roc_curve
# Random Forest Classification
Y_predict5_proba = rfcla.predict_proba(X_test)
Y_predict5_proba = Y_predict5_proba[:, 1]
fpr, tpr, thresholds = roc_curve(Y_test, Y_predict5_proba)
plt.subplot(331)
plt.plot([0,1],[0,1],'k--')
plt.plot(fpr,tpr, label='ANN')
plt.xlabel('fpr')
plt.ylabel('tpr')
plt.title('ROC Curve Random Forest')
plt.grid(True)
plt.subplots_adjust(top=2, bottom=0.08, left=0.10, right=1.4, hspace=0.45, wspace=0.45)
plt.show()
# Naive Bayes Classification
Y_predict3_proba = nbcla.predict_proba(X_test)
Y_predict3_proba = Y_predict3_proba[:, 1]
fpr, tpr, thresholds = roc_curve(Y_test, Y_predict3_proba)
plt.subplot(332)
plt.plot([0,1],[0,1],'k--')
plt.plot(fpr,tpr, label='ANN')
plt.xlabel('fpr')
plt.ylabel('tpr')
plt.title('ROC Curve Naive Bayes')
plt.grid(True)
plt.subplots_adjust(top=2, bottom=0.08, left=0.10, right=1.4, hspace=0.45, wspace=0.45)
plt.show()
Conclusion¶
Model Comparison & Practical Applications¶
Why Random Forest Outperformed Naive Bayes¶
Handles Complexity Better:
- RF captures interactions between features (e.g., how deposit patterns + customer service calls combine to predict churn)
- Naive Bayes assumes all features act independently (often too simplistic for real-world behavior)
More Robust with Real Data:
- Works well with both categorical (player tier) and numerical (deposit amounts) features
- Naturally handles missing values and outliers
- Provides feature importance scores (shows which factors matter most)
What This Model Enables¶
1. Real-Time Relationship Scoring¶
Deploy as a web app where staff can:
- Input player ID or upload data
- Instantly see:
- 🔴 High Risk (likely to churn)
- 🟡 Warning Signs (needs attention)
- 🟢 Healthy (stable relationship)
2. Automated Monitoring¶
# Sample prediction for new data
def predict_relationship(new_player_data):
churn_prob = rf_model.predict_proba(new_player_data)[:,1]
if churn_prob > 0.7:
return "🔴 High Risk - Immediate action suggested"
elif churn_prob > 0.4:
return "🟡 Warning - Monitor closely"
else:
return "🟢 Healthy - Standard engagement"
3. Targeted Retention¶
Flag at-risk players for:
- Personalized bonus offers (tailored to player history)
- VIP concierge outreach (dedicated account manager contact)
- Payment method updates (reduce failed transactions)
Next-Level Implementation¶
API Integration¶
- Connect directly to your CRM to:
- Score all players daily
- Update churn risk in real-time
Automated Alerts¶
- Configure notifications for:
- Email/Slack alerts when VIP players become high-risk
- Real-time dashboards for customer support teams
Root Cause Analysis¶
- Use feature importance to:
- Identify operational pain points (e.g., failed transactions = 23% of churn)
- Prioritize product improvements
"This isn't just academic - it's a working tool that turns data into retention strategies."