jameskingsley/mtn-nigeria-churn-prediction

nd-to-end customer churn prediction and analysis framework using MTN Nigeria's Q1 2025 dataset. Includes EDA, MLOps tracking (MLflow/ClearML), API deployment (FastAPI), interactive UI (Streamlit), and reporting via Looker Studio # MTN Nigeria Customer Churn Prediction Engine An enterprise-grade, end-to-end MLOps production pipeline that trains, registers, hosts, and serves an optimized machine learning model to predict subscriber churn for MTN Nigeria. This system completely decouples code from heavy binary artifacts, utilizing a centralized cloud model registry for secure, authenticated artifact streaming at runtime. --- ## Live Infrastructure Links * **Production API Gateway (FastAPI):** [mtn-nigeria-churn-predictio…] * **Interactive Scoring Portal (Streamlit):** [mtn-nigeria-churn-predictio…] * **Dashboard Link:** [datastudio.google.com] --- ## System Architecture & MLOps Workflow The system is split into three completely decoupled layers to maximize scalability, optimize system resources, and prevent repository bloat: 1. **Experimentation & Registration Layer (`src/models/train.py`):** Runs a localized, stratified model tournament across multiple candidate architectures (Random Forest, Gradient Boosting, Logistic Regression). The winning pipeline is versioned and its physical binary `.pkl` weight artifact is securely pushed up to ClearML Cloud Storage. 2. **Production Inference Layer (`src/main.py`):** A cloud-native FastAPI instance deployed on **Render**. On the initial bootstrap request, it securely authenticates via environment keys, downloads the active model weights from ClearML directly into memory cache via an authenticated stream, and processes high-throughput predictive queries. 3. **User Gateway Layer:** A lightweight Streamlit UI providing business stakeholders with an intuitive interface to query customer behavior patterns against the live cloud endpoint. --- ## Technology Stack * **Core Language:** Python * **Modeling & Pipelines:** Scikit-Learn, Pandas, NumPy, Joblib * **Experiment Tracking & Artifact Registry:** ClearML Cloud, MLflow (Local Tracking) * **Application Layer:* …