Project ID: 3CSE4_01

Synopsis

on

“Predict Whether a client will subscribe to a bank

term deposit”

Department of AIML

School of Computer Science and Engineering

IILM University, Greater Noida

Submitted By: Under the Supervision of:

1. Gaurav Kumar Singh (CS-23411167) Dr. Sumit Kumar
2. Devansh Kumar (CS-23411201) (Associate Professor)
3. Manasv Kaul (CS-23411287)

1
 Table of Contents
1. Abstract 3

2. Problem Statement 4

2. Objectives 5

3. Introduction 6

4. Literature Survey 6
4.1 Scope and Limitation 7

5. Gaps Identified 8

6. Proposed Methodology 9

6.1. DFD/Flow Chart or Architecture 10

6.2 Technology Used 13

7. Algorithms and Techniques 14

8. Conclusion and Future Scope 16

9. Reference 17

Appendix

2
Abstract
Direct marketing remains a crucial strategy for financial institutions, yet untargeted
campaigns, particularly for products like term deposits, often suffer from low conversion
rates and high operational costs. The primary challenge lies in efficiently identifying potential
clients who are most likely to subscribe. This project addresses this challenge by developing a
predictive model using machine learning techniques to forecast whether a client will
subscribe to a bank term deposit based on data from past marketing campaigns.

This study will utilize a public "Bank Marketing Dataset," which includes client demographic
data, information on past interactions, and broader economic indicators. The methodology
will involve comprehensive exploratory data analysis (EDA), data preprocessing (handling
missing values, encoding categorical variables), and feature engineering to identify the most
significant predictors of subscription.

Several supervised classification algorithms, such as Logistic Regression, Decision Trees,

Random Forest, and Gradient Boosting models, will be implemented and rigorously
evaluated. Given the typical class imbalance in such datasets (where non-subscribers far
outnumber subscribers), model performance will be assessed using metrics beyond simple
accuracy, including Precision, Recall, F1-Score, and the Area Under the ROC Curve (AUC).

The anticipated outcome is a robust and well-tuned classification model. This model will
serve as a powerful tool for the bank to segment its client base, enabling targeted marketing
efforts focused on high-propensity individuals. By optimizing its campaign strategy, the bank
can expect to see a significant improvement in conversion rates, a reduction in marketing
expenditure, and an overall enhancement of its marketing effectiveness.

Keywords: Machine Learning, Predictive Modeling, Bank Marketing, Term Deposit,

Customer-Propensity, Classification, Data Mining, Supervised Learning.

3
Problem Statement
In the highly competitive banking sector, the acquisition of term deposits is a key objective
for maintaining liquidity and funding operations. Banks traditionally rely on direct marketing
campaigns, such as telemarketing, to reach potential subscribers. However, these campaigns
are often executed with minimal targeting, resulting in several significant problems:

1. Low Conversion Rates: A large volume of calls is made to clients who have little to
no interest in the product. This "cold-calling" approach yields a very low success rate,
making the entire campaign highly inefficient.

2. High Operational Costs: Each contact made, whether successful or not, incurs a cost
in terms of human resources (call center staff), time, and operational overhead. An
untargeted campaign inflates these costs by spending resources on non-receptive
clients.

3. Customer-side Animosity (Marketing Fatigue): Unwanted and irrelevant marketing

calls can lead to customer dissatisfaction and annoyance. This "marketing fatigue" not
only damages the bank's relationship with the client but also reduces the effectiveness
of future marketing initiatives.

The core of the problem is the lack of a data-driven system to differentiate between clients
who are likely to subscribe and those who are not. The bank needs to move from a "one-size-
fits-all" mass-marketing strategy to a precise, targeted approach.

Therefore, this project aims to address this inefficiency by developing a predictive model.
The model will analyze historical client data and campaign outcomes to generate a
"propensity score" for each client, indicating their likelihood of subscribing to a term deposit.
This will enable the bank to focus its valuable marketing resources exclusively on high-
propensity clients, thereby increasing conversion rates, optimizing costs, and improving
customer relations.

4
Objectives
The primary objective of this project is to develop and evaluate a machine learning model
that accurately predicts the likelihood of a client subscribing to a bank term deposit.

To achieve this primary goal, the following specific objectives are defined:

1. To conduct Exploratory Data Analysis (EDA): This involves analyzing the bank
marketing dataset to understand data distributions, identify correlations between
variables (e.g., age, job, previous contact outcome), and visualize key relationships
that influence a client's subscription decision.

2. To pre-process and prepare the data: This objective involves all necessary data-
cleaning tasks, such as handling missing values, treating outliers, encoding categorical
variables into a numerical format, and performing feature scaling.

3. To perform feature engineering and selection: This objective is to create new

relevant features from existing data and to identify and select the most significant
subset of features that contribute most to the model's predictive power, helping to
reduce model complexity and improve performance.

4. To build and train classification models: To implement several supervised

classification algorithms (such as Logistic Regression, Decision Trees, Random
Forest, and Gradient Boosting) to learn patterns from the historical campaign data.

5. To evaluate and compare model performance: To assess the effectiveness of the

trained models using appropriate evaluation metrics. Given the imbalanced nature of
the dataset (few subscribers vs. many non-subscribers), the focus will be on metrics
like Precision, Recall, F1-Score, and the Area Under the ROC Curve (AUC) rather
than just accuracy.

6. To deliver actionable insights: To identify the best-performing model and interpret

its findings to provide the bank with a clear understanding of the key factors that drive
term deposit subscriptions.

5
Introduction
In today's data-driven world, the financial services industry possesses a vast amount of
customer data. The effective use of this data has become a critical differentiator for success.
For banks, direct marketing campaigns are a primary tool for acquiring customers for
products like term deposits. However, the effectiveness of these campaigns is often hindered
by a lack of precise targeting, leading to wasted resources and customer fatigue. This is the
central challenge this project seeks to address.

The emergence of machine learning has provided powerful tools for analyzing historical data
to identify complex patterns and make accurate predictions. By applying machine learning
techniques to past marketing data, it is possible to build a model that can predict, with a
reasonable degree of accuracy, which clients are most likely to respond positively to a
marketing offer.

This project will focus on developing such a predictive model. Using a dataset from a
Portuguese bank's telemarketing campaign, this study will systematically explore the data,
engineer relevant features, and build and compare several classification models. The ultimate
goal is to create a tool that can help the bank optimize its marketing strategy by identifying
and targeting a high-propensity cohort of clients, thereby increasing the campaign's return on
investment (ROI) and enhancing customer relationships.

This synopsis outlines the project's core components. It begins by defining the problem and
its business implications, followed by a clear set of objectives. It then reviews existing
literature in the field, details the proposed methodology, and discusses the project's scope and
inherent limitations.

Literature Survey
The application of data mining and machine learning in the financial domain is a well-
established field of study. A significant body of research focuses on customer relationship
management (CRM), churn prediction, and, as in this case, marketing campaign optimization.

Previous studies on the same or similar bank marketing datasets (Moro, et. al., 2014) have
consistently demonstrated the superiority of non-linear classification models over traditional

6
statistical methods. Researchers have frequently employed algorithms like Decision Trees,
which provide human-readable rules and highlight key decision factors (e.g., "if 'duration' >
X and 'poutcome' = 'success', then 'subscribe' = 'yes'"). Support Vector Machines (SVMs) and
Neural Networks have also been applied, often yielding high accuracy but at the cost of
interpretability.

More recent studies emphasize ensemble methods, such as Random Forest and Gradient
Boosting Machines (e.g., XGBoost, LightGBM), as they typically achieve the highest
predictive performance by combining multiple weak learners. Another critical theme in the
literature is the problem of class imbalance. Since subscribers are a small minority in the
data, standard models tend to be biased toward the majority "non-subscriber" class. Common
solutions explored include undersampling the majority class, oversampling the minority class
(e.g., using SMOTE - Synthetic Minority Over-sampling Technique), or using cost-sensitive
learning.

This project will build upon this existing body of work by implementing and comparing
several of these state-of-the-art models, with a particular focus on handling class imbalance
and ensuring the final model is not only accurate but also interpretable for business
stakeholders.

4.1. Scope and Limitations

The project focuses on building a supervised classification model using historical client and
campaign data from the banking sector. The scope includes data preprocessing, feature
engineering, model training, evaluation, and interpretation of results. The system can assist
banks in identifying potential customers and optimizing campaign strategies. However,
limitations include reliance on the quality and completeness of the dataset, potential biases in
the data, and the fact that customer decisions may also be influenced by external factors (such
as economic conditions or personal circumstances) not captured in the dataset

Gaps Identified
Based on the literature survey, several gaps in existing research have been identified that this
project aims to address:

1. The "Call Duration" Paradox: Many studies (e.g., Moro et. al., 2014) acknowledge
that the 'duration' of the call is the single most dominant predictor of subscription.

7
 However, this feature is unknown before a call is made, rendering it useless for a pre-
campaign targeting model. A significant gap exists in studies that build models after
explicitly excluding this feature, which is the only practical approach for client-
selection. This project will focus on building a model without this variable, simulating
a real-world targeting scenario.

2. Interpretability of Complex Models: While ensemble models (Random Forest,

Gradient Boosting) are known to be the most accurate, they are often treated as "black
boxes." This creates a gap for business users who need to understand why a client is
flagged as a high-propensity lead. This project will attempt to bridge this gap by
applying post-hoc interpretability techniques (like SHAP values or feature importance
plots) to the best-performing complex model.

3. Limited Scope of Feature Engineering: Many studies use the provided features as-
is. There is a gap in exploring more creative or domain-specific feature engineering.
For example, creating features that represent a client's "economic engagement" (e.g.,
loan status vs. housing) or "communication fatigue" (e.g., ratio of contacts in this
campaign vs. previous campaigns) could yield new insights.

By focusing on these practical gaps, this project aims to produce a model that is not only
statistically sound but also genuinely useful and deployable in a real-world business context.