Customer Churn Prediction

1. Introduction

The goal of this project is to prepare data for customer churn prediction. We checked the data, did EDA (Exploratory Data Analysis), and suggested preprocessing steps.

Dataset description:

• 1000 rows

• 5 columns:

  • CustomerID – unique ID for each customer

  • Age – customer age (18–65)

  • Gender – M or F

  • MaritalStatus – Single, Married, Divorced, Widowed

  • IncomeLevel – Low, Medium, High

---

2. Exploratory Data Analysis (EDA)

2.1 Age Distribution

• Customer ages are mainly between 18 and 65.

• The age distribution is quite even, but some ages look too frequent (synthetic data sign).

2.2 Gender Distribution

• Gender is almost equal (balanced).

2.3 MaritalStatus vs IncomeLevel

IncomeLevel	Divorced	Married	Single	Widowed
High	82	98	75	94
Low	87	86	62	90
Medium	79	77	78	92

Observation:

• High and Medium income are evenly spread in all marital groups.

• Low income is similar but Single customers are fewer.

2.4 MaritalStatus vs AgeGroup

AgeGroup	Divorced	Married	Single	Widowed
<18	4	8	7	5
18–35	80	84	69	93
36–50	67	76	63	77
51–65	76	67	64	76
65+	21	26	12	25

Observation:

• Age group 18–35 is the largest in all marital groups.

• Widowed is more common in young people (synthetic data).

• Age 65+ has mostly Married and Widowed.

2.5 IncomeLevel vs AgeGroup

AgeGroup	High	Low	Medium
<18	4	10	10
18–35	121	105	100
36–50	94	98	91
51–65	105	86	92
65+	25	26	33
Observation:

• High income is mainly 18–35 (121 people).

• Low and Medium incomes are balanced across ages.

• Age 65+ mostly has Medium income.

---

3. Data Cleaning and Preprocessing

• Missing values: None.

• Outliers: Age has no extreme values (range 18–65).

• Feature Engineering: Added AgeGroup (<18, 18–35, 36–50, 51–65, 65+).

• Encoding:

  • Gender → Binary (M=1, F=0)

  • MaritalStatus, IncomeLevel, AgeGroup → One-hot encoding

• Normalization: Age can be scaled using StandardScaler or MinMaxScaler.

• Redundant Columns: CustomerID is not needed for the model.

---

4. Findings (Main observations)

• Age is strongly related to IncomeLevel and MaritalStatus.

• Age 18–35 dominates both marital status and high income groups.

• Widowed in young people and many young high-income customers are signs of synthetic data.

• Dataset does not fully match real-world patterns but is useful for model preparation.