Telco Customer Churn Prediction
Overview:
This project focuses on predicting customer churn for a telecommunications company using Support Vector Machine (SVM) models. A key achievement of this project is the development of an SVM model from scratch, which demonstrated a 0.4% improvement in accuracy over a pre-built SVM model from standard machine learning libraries.
Project Highlights:
Custom SVM Model: Implemented an SVM model from scratch, achieving superior performance compared to a pre-built SVM model.
• Data Preprocessing:
Applied label encoding for categorical features. Standardized features using StandardScaler. Addressed missing values by imputing with the median. Replaced inconsistent values for uniformity.
• Data Visualization:
Explored the relationship between various services and churn, as well as gender-specific churn patterns. Workflow:
• Data Cleaning:
Handled missing values by replacing them with the median. Standardized text values to ensure consistency.
• Data Splitting:
Divided the dataset into 80% training and 20% testing sets to assess model performance. Model Development:
• Custom SVM Model:
Developed from scratch, with fine-tuning to maximize performance.
• Pre-built SVM Model:
Utilized a pre-built SVM model from a standard machine learning library for benchmarking.
• Model Evaluation:
Evaluated both models using accuracy score, classification report, and confusion matrix. The custom SVM model achieved a 0.4% higher accuracy compared to the pre-built model.
• Data Visualization:
Conducted detailed visual analysis of service usage and churn rates. Analyzed churn rates by gender. Results
• Accuracy:
The custom SVM model demonstrated a 0.4% improvement in accuracy over the pre-built SVM model, showcasing the effectiveness of the custom implementation.
• Evaluation Metrics:
Comprehensive evaluation included accuracy score, classification report, and confusion matrix to assess model performance.
•Contribution and Support:
If you find any areas for improvement or have suggestions, feel free to request a commit on GitHub. For any questions or further discussion, you can reach out to me on the social media platforms mentioned in my profile.
•Conclusion;
This project underscores the potential of custom model development in enhancing predictive performance. It also provides a thorough approach to data preprocessing, model evaluation, and visualization for machine learning applications.