How pradnya1208/telecom-customer-churn-prediction works
Project type and purpose
• This is a data‐science/ML project implemented as a Jupyter notebook (“Customer churn prediction.ipynb”) under the Scripts folder.
• Its purpose is to ingest a telecom‐churn dataset, perform exploratory data analysis (EDA), train multiple classification models, evaluate them (with k-fold CV and confusion matrices), and produce visual output in the output/ directory.
File structure
• Top level:
• README.md – project overview, objectives, results.
• Scripts/ – the core code and raw data:
– Customer churn prediction.ipynb – the main notebook orchestrating ingestion → EDA → modeling → evaluation.
– data.csv – the raw telecom‐churn dataset.
• output/ – generated charts, images, confusion matrices, model‐comparison plots.
• No traditional “backend” or “frontend,” but a linear data‐processing pipeline in notebook form.
README for additional insights
• It describes:
• Dataset attributes (customer demographics, services, tenure, charges).
• EDA steps and key findings (distribution plots).
• ML models used (Logistic Regression, KNN, Naive Bayes, Decision Tree, Random Forest, AdaBoost, Gradient Boosting, Voting Classifier).
• Final model selection and confusion matrix stats.
a. Main components
1. Data Source
– Kaggle Telco Customer Churn CSV (data.csv under Scripts/)
2. Data Processing & Feature Engineering
– pandas data‐ cleaning, handling missing values, encoding categorical features.
3. Exploratory Data Analysis (EDA)
– seaborn/matplotlib steps that produce distribution plots, correlation heatmaps, service‐level comparisons.
4. Model Training Pipeline
– individual model trainers (LR, KNN, NB, DT, RF, AdaBoost, GradBoost)
– k-fold cross validation loop
– hyperparameter tuning placeholder
5. Ensemble / Voting Classifier
– combines gradient boosting, logistic regression, AdaBoost
6. Model Evaluation & Output
– accuracy scores, confusion matrices, saved plots → output/
7. Environment & Tools
– Jupyter Notebook, Python interpreter, scikit-learn, pandas, seaborn, NumPy
b. Relationships & Data Flow
– From processed features → each Model Training component (parallel branches)
– From each model’s CV results → Model Comparison block
– From top three models → Voting Classifier block → Final predictions
– From Final predictions → Evaluation block → Outputs
c. Architectural patterns & design principles
– Batch‐processing pipeline (one‐shot notebook run)
– Modular training steps per algorithm (separate stages)
– Ensemble pattern for final classifier
– Clear separation between data ingestion, analysis, training, and evaluation
d. Key technologies & libraries
– Python (3.x), Jupyter Notebook
– pandas, NumPy (data wrangling)
– seaborn, matplotlib (visualization)
– scikit-learn (modeling: classifiers, cross‐validation, ensemble)