ugly-custard / codsoft

Project-based internship projects

Codsoft internship projects

This repo lists the projects I worked on while doing the internship for Codsoft.

The projects are:

• Customer Churn Prediction
• Movie Genre Classification
• SMS Spam Detection

Customer Churn Prediction

The Dataset used: https://www.kaggle.com/datasets/shantanudhakadd/bank-customer-churn-prediction

In this project, using the powerful Random Forest Classifier, we leveraged historical data like usage behavior and demographics of customers to predict customer churn. Initially, the model achieved an impressive 0.87 accuracy. However, when addressing class imbalance via random under-sampling, accuracy dipped to 0.76, highlighting the balance challenge. In contrast, random over-sampling boosted accuracy to a remarkable 0.94! The feature analysis identified Age as the top predictor, followed by Balance, Estimated Salary, and Credit Score, in that order. These insights underscore their pivotal roles in predicting customer churn. In summary, the analysis of customer churn prediction yielded valuable insights. The choice between under-sampling and over-sampling depends on the balance- accuracy trade-off, with these key features playing crucial roles in customer retention analysis.

Movie Genre Classification

The Dataset used: https://www.kaggle.com/datasets/hijest/genre-classification-dataset-imdb

In this project, I compared three popular classifiers – SGD, Multinomial NB, and Logistic Regression – to classify movies into 27 different genres based on the movie synopsis. The results show that Logistic Regression outperforms the other models for this task. With an accuracy of 0.60, precision at 0.58, recall of 0.60, and a solid F1-score of 0.56, it consistently showcases balanced performance across various metrics. This means the Logistic Regression model effectively handles the complexities of the data, making it the optimal choice for this multi-class classification task.

SMS Spam Detection

The Dataset used: https://www.kaggle.com/datasets/uciml/sms-spam-collection-dataset

In this project, I put the spotlight on two trusty classifiers – SGD and Multinomial NB – to tackle the task of identifying spam messages with precision. The testing and evaluation reveal the following verdict: both the SGD Classifier and Multinomial NB models delivered exceptional performance! With identical high scores of 0.96 for accuracy, precision, recall, and F1-score, it's safe to say that these models are equally well-equipped for the task. In summary, the SGD Classifier and Multinomial NB models stand shoulder to shoulder, showcasing excellent classification capabilities and proving their mettle in SMS spam detection.

These are projects I have worked on for the project-based Machine Learning Internship by Codsoft.