How geethasagarb/coursera-churn-prediction-challenge works
Main Components
• Data Sources: Include the raw CSV files (train.csv, test.csv, data_descriptions.csv) and the output file (prediction_submission.csv) which represent the input datasets and the predicted results.
• Jupyter Notebook Environment: The ChurnPrediction.ipynb acts as the central orchestrator of the machine learning workflow.
• Data Cleaning & Preprocessing Module: Represents steps for handling missing values, converting categorical variables to numeric, and applying encoding techniques (e.g., One-Hot Encoding via ColumnTransformer).
• Feature Engineering Module: Emphasis on the creation of additional features derived from subscription details, usage behavior, etc.
• Model Training & Evaluation: The core machine learning component where a Logistic Regression model is built inside a pipeline, with configuration for hyperparameters and evaluation using ROC AUC score.
• Business Outcome Component: Represents the insights generated for proactive churn intervention, emphasizing cost reduction and retention strategies.
Relationships and Interactions
• Data Flow: Start from the data sources, then show the sequential flow into preprocessing, feature engineering, model training, and finally model evaluation and prediction output.
• Pipelines: Highlight the pipeline-based approach which integrates preprocessing and model training, reinforcing the modular and sequential architecture.
Architectural Patterns & Design Principles
• Pipeline Architecture: Central to integrating multiple stages (preprocessing, feature engineering, and logistic regression) in a streamlined flow.
• Separation of Concerns: Distinct modules handling data input, transformation, model development, and evaluation.
• Reproducibility & Scalability: The notebook-based workflow coupled with modular code encourages easy updates, testing, and potential deployment (as an API or integration into a CRM system).
Relevant Technologies and Frameworks
• Python, Jupyter Notebook, Pandas, NumPy, and Scikit-learn (especially for model development, feature engineering via ColumnTransformer, and logistic regression).
• Annotation: For each component, include brief notes about the key functions, such as “One-Hot Encoding using ColumnTransformer” in Preprocessing, or “Logistic Regression with Hyperparameter Settings” in Model Training.
• Detail Data Flow: Illustrate that train.csv feeds into the preprocessing module, while test.csv also goes through similar processing to generate predicted outputs stored in prediction_submission.csv.