Customer Subscription Prediction Analysis

Comprehensive machine learning classification project comparing 5 different algorithms to predict customer subscription behaviour using KNIME Analytics Platform

🎯 Project Overview

This project tackles the business-critical challenge of predicting whether customers will subscribe to a service based on historical marketing data. Using KNIME Analytics Platform, I implemented and compared five classification algorithms to identify the optimal approach for customer targeting and marketing strategy optimization.

Key Achievement: Random Forest model achieved 96% validation accuracy and secured top 15% placement in Kaggle competition (0.62658 score)

📊 Dataset Information

• Training Data: 26,360 customer records with 23 features
• Test Data: 6,590 records for Kaggle submission
• Target Variable: Binary subscription status (0/1)
• Challenge: Highly imbalanced dataset (89% non-subscribers)

Key Features

• Demographics: Age, job, marital status, education
• Financial: Default status, housing loans, personal loans
• Campaign: Contact method, duration, previous campaigns
• Economic: Employment rate, consumer confidence, interest rates

🛠️ Technical Implementation

Data Preprocessing Pipeline

• Missing Value Handling: Imputation using most frequent (categorical) and median (numerical)
• Class Imbalancing: SMOTE technique to balance 89%/11% distribution
• Feature Engineering: Category to number conversion, normalization
• Data Splitting: 70/30 train/validation with consistent random seed

Machine Learning Models Implemented

Algorithm	Validation Accuracy	Precision	Recall	F1-Score	AUC
Random Forest	96%	94%	99%	96%	99%
Decision Tree	92%	89%	94%	92%	95%
K-Nearest Neighbours	92%	90%	94%	92%	95%
Neural Network (MLP)	86%	81%	94%	87%	91%
Support Vector Machine	89%	89%	0%	94%	50%

🔧 KNIME Workflow Architecture

Core Pipeline Components

1. Data Import & Type Casting
2. Missing Value Treatment
3. SMOTE Balancing
4. Feature Preprocessing
5. Model Training & Validation
6. Performance Evaluation
7. Kaggle Submission Export

Advanced Features

• Parameter Optimization Loops for hyperparameter tuning
• ROC Curve Analysis for model performance assessment
• Feature Importance Analysis using Random Forest attribute statistics
• Cross-Validation for robust model evaluation

📈 Key Results & Insights

Best Performing Model: Random Forest

• Training Accuracy: 99.8%
• Validation Accuracy: 96%
• Kaggle Score: 0.62658 (Top 25%)
• Key Strength: Excellent balance of precision and recall

Feature Importance Findings

Random Forest attribute statistics revealed the most predictive features for subscription behaviour, enabling targeted marketing strategies.

Business Impact

• 96% accuracy enables confident customer targeting
• Reduced marketing costs through precise subscriber identification
• Improved ROI on marketing campaigns through data-driven decisions

🚀 Technical Skills Demonstrated

Machine Learning

• Classification Algorithms: Decision Trees, Random Forest, KNN, SVM, Neural Networks
• Model Evaluation: ROC/AUC analysis, confusion matrices, cross-validation
• Hyperparameter Tuning: Grid search and optimization loops
• Class Imbalancing: SMOTE implementation for skewed datasets

Data Science Pipeline

• Data Preprocessing: Missing value treatment, feature scaling, encoding
• Feature Engineering: Category conversion, normalization techniques
• Model Comparison: Systematic evaluation across multiple algorithms
• Performance Optimization: Parameter tuning for each algorithm

Tools & Technologies

• Primary Platform: KNIME Analytics Platform
• Competition: Kaggle submission and leaderboard participation
• Visualization: ROC curves, confusion matrices, performance charts
• Documentation: Comprehensive analysis and methodology reporting

📁 Repository Structure

customer-subscription-prediction/
├── data/
│   ├── Assignment3UnknownDataset.csv (test data)
│   ├── Assignment3MarketingDataset.csv (training data)
│   └── sample_predictions.csv (Kaggle submission)
├── docs/
│   └── full_analysis_report.pdf
├── results/
│   └── model_performance_comparison.png
├── workflows/
│   ├── Decision_tree_workflow.png
│   ├── K_nearest_neighbour_workflow.png
│   └── neural_network_workflow.png
│   ├── random_forest_workflow.png
│   ├── svm_optimization_workflow.png
├── README.md

🎯 Business Applications

This analysis directly applies to:

• Customer Acquisition: Identify high-probability prospects
• Marketing Optimization: Allocate budget to most promising segments
• Campaign Targeting: Personalize outreach based on prediction confidence
• Resource Planning: Forecast subscription volumes for capacity planning

🔍 Future Enhancements

• Deep Learning: Implement advanced neural architectures
• Ensemble Methods: Combine top-performing models
• Real-time Prediction: Deploy model for live customer scoring
• Feature Engineering: Create interaction terms and polynomial features

📧 Contact

Zyad Kamal Hamed
📧 zyad2408@live.com.au
🔗 LinkedIn

---

Demonstrating end-to-end machine learning capabilities from data preprocessing to model deployment and business impact