Scribd
Upload
6 views9 pages

Phase-2 Ibrahim

The document outlines a project focused on predicting customer churn in the telecommunications industry using machine learning techniques. It details the problem statement, project objectives, data description, preprocessing steps, exploratory data analysis, model building, and team contributions. The project aims to develop a classification model to identify churn likelihood and improve customer retention strategies.

Uploaded by

azhanmohammed04
Copyright
© © All Rights Reserved
We take content rights seriously. If you suspect this is your content, claim it here.
Available Formats
Download as PDF, TXT or read online on Scribd
6 views9 pages

Phase-2 Ibrahim

The document outlines a project focused on predicting customer churn in the telecommunications industry using machine learning techniques. It details the problem statement, project objectives, data description, preprocessing steps, exploratory data analysis, model building, and team contributions. The project aims to develop a classification model to identify churn likelihood and improve customer retention strategies.

Uploaded by

azhanmohammed04
Copyright
© © All Rights Reserved
We take content rights seriously. If you suspect this is your content, claim it here.
Available Formats
Download as PDF, TXT or read online on Scribd
You are on page 1/ 9

Phase-2 Submission Template

Student Name: MOHAMMED IBRAHIM

Register Number: 510623104058


Institution: C ABDUL HAKEEM COLEGE OF ENGINEERING
& TECHNOLOGY
Department: COMPUTER SCIENCE & ENGINEERING

Date of Submission: 08-05-2025


Github Repository Link: https://github.com/ibrahim5826/customer-
support-chatbot.git

1. Problem Statement

Customer churn is a critical metric for businesses, especially in highly competitive


industries like telecommunications. Churn refers to the rate at which customers stop
doing business with a company. Our project aims to predict whether a customer is
likely to churn based on their service usage patterns, contract details, and
demographics.
• Refinement from Phase-1:
Initially, the problem was understood in broad terms. After analyzing the
dataset, we realized that several categorical variables (e.g., Contract,
PaymentMethod, InternetService) significantly influence churn behavior.
Therefore, we narrowed our focus to predicting churn using a classification
model.
• Problem Type: Binary Classification The target variable (Churn) has two
possible outcomes: "Yes" or "No".
• Why it matters:
Accurately predicting churn can help businesses reduce customer loss by
targeting retention strategies, leading to improved customer satisfaction and
revenue stability.

2. Project Objectives
• Build machine learning models that can classify whether a customer will
churn.

• Achieve high model accuracy and balanced precision-recall to handle class


imbalance.

• Identify the most influential features contributing to churn.

• Make model outputs interpretable for business stakeholders.

Updated Objective: After EDA, we included feature importance analysis and


customer profiling to better understand churn causes.
3. Flowchart of the Project Workflow

Data Collection

Data Cleaning and Processing

Exploratory Data Analysis(EDA)

Future Engineering

Model Building and Evaluation

Result Interpretation & Visualization

Documentation & Deployment Plan

4. Data Description

• Dataset Name: Telco Customer Churn Dataset

• Source: Kaggle - IBM Sample Dataset

• Type of Data: Structured tabular data

• Number of Records: 7,043 rows

• Number of Features: 21 features (excluding customer ID)

• Static/Dynamic: Static snapshot

• Target Variable: Churn (Yes/No)


5. Data Preprocessing

• Missing Values:

o Column TotalCharges had 11 missing values due to blank entries.


These were imputed using the median value of the column.

• Duplicate Records:

o Checked using df.duplicated().sum() → Result: 0 duplicates.

• Outliers:

o Outliers in MonthlyCharges and TotalCharges were identified using


boxplots. Handled using winsorization for extreme cases.

• Data Type Conversion:

o TotalCharges was originally an object type. Converted to float using


pd.to_numeric().

• Categorical Encoding:

o Binary columns (e.g., gender, Partner) were label encoded.

o Multi-category columns (e.g., PaymentMethod, InternetService) were


one-hot encoded.

• Feature Scaling:

o Numerical features (tenure, MonthlyCharges, TotalCharges) were


standardized using StandardScaler.
6. Exploratory Data Analysis (EDA)

• Univariate Analysis:

o Churn: 26.5% customers churned (imbalanced target) o Contract:

Most churn occurs in month-to-month contracts o Visuals used:

Histograms, boxplots, countplots

• Bivariate/Multivariate Analysis:

o Correlation matrix: Showed strong correlation between tenure and

MonthlyCharges with churn o Pairplots and groupby plots:

▪ Customers with fiber optic internet churn more often ▪

Customers using electronic checks are more likely to churn

• Insights Summary:

o tenure is inversely related to churn o Longer contract types (1-year or

2-year) have lower churn rates o Services like tech support and online

backup seem to retain customers

7. Feature Engineering
New Features Created:
• TenureGroup: Categorized tenure into "0–12", "12–24", etc.

• HasMultipleServices: Combined multiple service features to count total


services per customer

Transformed Features:

• Created interaction terms like MonthlyCharges * Tenure

Dimensionality Reduction:

• PCA was attempted but didn't improve model performance significantly, so


not retained in final model.

Feature Selection:

• Used Recursive Feature Elimination (RFE) to choose top 10 important


features

8. Model Building

• Train/Test Split:
o 80% training, 20% testing; stratified on target to maintain class
balance

• Models Implemented:
o Logistic Regression: Baseline model
o Random Forest Classifier: Non-linear model for improved
performance
• Evaluation Metrics:
o Accuracy, Precision, Recall, F1-Score, ROC AUC
F1-
Model Accuracy Precision Recall AUC

Logistic Regression 0.80 0.71 0.67 0.69 0.83


Random Forest 0.86 0.78 0.76 0.77 0.89

9. Visualization of Results & Model Insights

• Confusion Matrix: Showed reduction in false positives for Random Forest

• ROC Curve: Random Forest had an AUC of 0.89 indicating strong


performance

• Feature Importance Plot: Contract, tenure, and PaymentMethod were most


influential

• Churn Profile Visuals: Created churn heatmaps by contract and tenure


group

10. Tools and Technologies Used

Programming Language: Python

Notebook Environment: Google Colab Libraries

Used:

• pandas, numpy – Data processing

• matplotlib, seaborn, plotly – Visualization

• scikit-learn – Modeling and evaluation


• xgboost (optional experiment)

11. Team Members and Contributions

Mohammed Musaddiq. M [510623104059]-Project Lead & Problem


Definition Responsible for defining the problem statement, coordinating
tasks, and ensuring the project follows the timeline. Oversees final
documentation and submission.

Mohammed Ammar Saqib [510623104055] - Data Collection & Cleaning


Gathers relevant datasets from public sources or generates synthetic data.
Handles data preprocessing (cleaning, formatting, normalization).

Ghani Adnan Faiz [510623104005] - Exploratory Data Analysis (EDA)


Analyzes data to uncover patterns and insights. Creates visualizations using
matplotlib/seaborn/plotly.

Fateh Mohammed [510623104024] - Feature Engineering & Model Building


Designs features, selects and trains models (e.g., intent classifiers, response
generators). Chooses appropriate NLP techniques.

Abraar. A [510623104003] - Model Evaluation & Interpretation


Evaluates model performance using metrics (accuracy, F1 score, etc.).
Prepares interpretation reports and validation results.
Mohammed Ibrahim – [510623104058] - Deployment & Frontend
Integration Builds and deploys the chatbot using Streamlit/Gradio/Flask.
Handles web interface design and chatbot testing.

You might also like

pFad - Phonifier reborn

Pfad - The Proxy pFad of © 2024 Garber Painting. All rights reserved.

Note: This service is not intended for secure transactions such as banking, social media, email, or purchasing. Use at your own risk. We assume no liability whatsoever for broken pages.


Alternative Proxies:

Alternative Proxy

pFad Proxy

pFad v3 Proxy

pFad v4 Proxy