Customer Churn Prediction – A Basic ML Pipeline

This video walks through a basic, baseline machine learning pipeline for customer churn prediction. The goal here is not to build the best possible model, but to establish a clean, correct foundation that we can improve step by step in future advanced versions. Using the Telco Customer Churn dataset (IBM / Hugging Face), we go from raw data to a working Random Forest model: - Initial data inspection & common pitfalls - Numeric columns hidden as text - Categorical encoding with one-hot encoding - Train-test split with stratification - Handling class imbalance with class_weight=balanced - Evaluating churn models beyond accuracy ⏱️ Timestamps: 00:00 Intro – Can We Predict Customer Churn? 00:42 Dataset Overview (Telco Churn – IBM / Hugging Face) 01:13 Data Inspection & Key Issues 04:28 Data Cleaning & Preprocessing 06:50 Encoding & Train-Test Split 09:59 Training the Random Forest Model 11:21 Model Evaluation & Recall Problem 14:27 Accuracy vs Recall – Business Takeaway ⚠️ Important note: This baseline model has low recall, meaning it misses many churners. That limitation is intentional — it sets the stage for advanced techniques like: - Threshold tuning - Alternative metrics - Model and feature improvements In later videos, we will deliberately build lower-accuracy models to achieve much higher recall, which is often the correct goal in real business settings. Think of this video as Churn Prediction – Level 1. 👉 If you need a clean and complete notebook with a professional LaTeX-ready report, feel free to check out the product at this link: https://mapml.gumroad.com/l/ml-churn-baseline #churnprediction #machinelearningproject #datascienceproject #telcocustomerchurn