---

Introduction

This project explores the performance of various classifiers—K Nearest Neighbor, Logistic Regression, Decision Trees, and Support Vector Machines (SVM)—to predict whether a client will subscribe to a term deposit using the Bank Marketing Dataset.

You can check out work out at: https://github.com/kushchhabra0/BankMarketingClassifer

---

Dataset

We use the Bank Marketing Dataset from the UCI Machine Learning Repository.

This dataset contains data related to direct marketing campaigns of a Portuguese banking institution. The goal is to predict whether a client will subscribe to a term deposit.

---

Usage

1. Load the dataset:

   

2. Open and run Bank_Marketing_Classifier_Comparison.ipynb.

3. Follow step-by-step:

   • Preprocessing

   • Model building

   • Evaluation

   • Visualization

---

Data Understanding and Preprocessing

Feature Engineering:

• Handle missing values

• Encode categorical variables

• Scale numerical features

Train/Test Split:

Split dataset into training and testing subsets.

Baseline Model:

Use the most frequent class to set a baseline accuracy.

Logistic Regression:

Implement logistic regression and evaluate its performance.

Model Comparisons:

Compare Logistic Regression, KNN, Decision Tree, and SVM using:

• Accuracy

• Training time

---

Visualizations

To understand relationships in the dataset, we used:

• Age distribution plot

  

• Box plot of Age by Subscription Status

  

• Categorical plots:

  • Job distribution

    

  • Marital status

    

  • Education levels

    

  • Subscription rates by category

    

---

Model Comparison

We compare the performance of:

• K Nearest Neighbors (KNN)

• Logistic Regression

• Decision Tree

• Support Vector Machine (SVM)

  ⚠️ Note: It takes approximately 5–7 minutes to run the model comparison part.

---

Outcomes

Baseline Accuracy: 0.8876

Logistic Regression

• Train Accuracy: 0.9013

• Test Accuracy: 0.8971

Best Results

• Best model overall: Logistic Regression

• Best tuned Decision Tree achieved:

  • Max Depth: 5

  • Min Samples Split: 2

  • CV Score: 0.9012

  • Test Accuracy: 0.8967

Improving the Model

We used hyperparameter tuning to improve model performance:

• Best parameters found:
  {'classifier__max_depth': 5, 'classifier__min_samples_split': 2}

• Best cross-validation score:
  0.9012

• Best Decision Tree Test Accuracy after tuning:
  0.8967

Conclusion

In this project, we evaluated several machine learning models to solve a binary classification problem using the Bank Marketing Dataset. After preprocessing, visual exploration, and modeling:

• Logistic Regression delivered the best balance between accuracy and training time.

• SVM performed similarly in terms of test accuracy but took significantly longer to train.

• Decision Trees, while achieving high training accuracy, suffered from overfitting.

Through hyperparameter tuning, the Decision Tree model's test accuracy significantly improved and approached Logistic Regression performance.

→ This shows the importance of model tuning and choosing the right algorithm based on dataset characteristics and computational resources.

Project Members

Kavish Jain (23UCS614)

Kshitij Sharma (23UCS628)

Kushal Chhabra(23UCS630)

---

---