Comparing Classifiers Notebook

Overview

This notebook focuses on a practical application of comparing the performance of several classifiers, including:

• K Nearest Neighbor (KNN)
• Logistic Regression (LogReg)
• Decision Trees (DT)
• Support Vector Machines (SVM)
• Naive Bayes (NB)

The data used is related to marketing bank products over the telephone and is sourced from the UCI Machine Learning repository. The dataset contains results from multiple marketing campaigns conducted by a Portuguese banking institution.

Contents

1. Understanding the Data

   • Dive deep into the data by reading the information provided in the UCI repository.
   • Understand the context and details of the marketing campaigns.

2. Data Loading and Exploration

   • Read the dataset bank-additional-full.csv using pandas.
   • Understand the features, including their data types and potential missing values.

3. Model Comparison

   • Train models with their default values and compare their performances.
   • Utilize GridSearchCV for hyperparameter tuning and observe the changes in model performance.

4. Hyperparameters

   • Detailed list of hyperparameters used for tuning each classifier.

Results

Performance with Default Parameters

Model	Train Time (sec)	Train Accuracy	Test Accuracy
LogReg	0.091	0.799	0.814
KNN	0.020	0.817	0.768
DT	0.052	0.900	0.657
SVM	6.358	0.802	0.814
NB	0.027	0.330	0.322

Performance with Hyperparameter Tuning (GridSearchCV)

Model	Train Time (sec)	Train Accuracy	Test Accuracy	Best Estimator
LogReg	2.965	0.799	0.814	{'LogReg__C': 1, 'LogReg__solver': 'newton-cg'}
KNN	2.801	0.830	0.735	{'KNN__metric': 'euclidean', ...}
DT	0.288	0.810	0.797	{'DT__max_depth': 10, 'DT__max_features': 'sqrt'}
SVM	15.545	0.721	0.726	{'SVM__C': 1, 'SVM__kernel': 'sigmoid'}

---