S-909/A-Machine-Learning-Approach-to-ISP-Churn-Prediction-in-Kenya

# A Machine Learning Approach to ISP Churn Prediction in Kenya ## Dataset This project uses a CSV dataset containing 36,992 customer records, each with 23 features. ## Key Target Variable The target variable is **'churn'**, indicating whether a customer has churned (Yes/No), defined as 30 days of inactivity or cancellation. ## Objectives - Predict customer churn risk with a recall of at least 70% to accurately identify high-risk customers. - Identify the main factors driving churn at both global and segment levels. - Integrate customer feedback and sentiment analysis into the predictive features. - Propose actionable strategies to reduce churn by at least 10% year-over-year. ## Business Benefits - **Lower Acquisition Costs:** Retaining existing customers is approximately five times cheaper than acquiring new ones. - **Increased Customer Lifetime Value (CLV):** Targeted offers help extend customer loyalty and lifetime. - **Customer-Centric Innovation:** Analyzing complaints and sentiment drives improvements in products and services. - **Revenue Stability:** Early detection of churn enables proactive retention efforts. # Methodology ## 1. Business Understanding - Defined churn as customers inactive for 30+ days or who have canceled. - Identified key performance indicators: retention rate and customer lifetime value (CLV). - Set success criteria: the predictive model must achieve at least 70% recall on churned customers. ## 2. Data Understanding - Conducted descriptive statistics to analyze data distribution and detect outliers. - Performed missing value analysis to identify and address data gaps. - Merged all datasets using a unified `customer_id` key to ensure consistency. ## 3. Data Preprocessing & Feature Engineering - **Missing Values:** Dropped less than 1.7% of rows due to missing data, which was negligible. - **Standardization:** Centered and scaled continuous variables for uniformity. - **Normalization:** Rescaled skewed features such as average sessio …