Pysham0n/Sales-Forecast-MlOps

In this project, we developed a multivariate forecasting solution to predict sales for a major Tunisian company. The dataset was initially unstructured, stored in scattered Word documents, requiring extensive data cleansing, transformation, and normalization to prepare it for analysis. # 📊 Problem Overview The original dataset was unstructured and scattered across multiple Word documents. This required: - 📌 **Data cleansing and transformation** - 📌 **Normalization and unification** for analysis-ready formatting --- # Workflow # 🧠 Multivariate Sales Forecasting with Full MLOps Pipeline ## 🔍 Time Series Analysis We performed a full time series analysis, including: - 📈 **Stationarity testing** - 🔁 **Trend and seasonality detection** We experimented with multiple **Machine Learning** and **Statistical models**: - **Support Vector Machine (SVM)** - **Linear Regression** - **Vector Auto Regression (VAR)** --- ## 🚀 MLOps Pipeline Implementation Once the optimal model was selected, we built and deployed a **full MLOps pipeline**, as illustrated in the diagram. ### 🔄 Data Versioning & Experiment Tracking - `DVC` - `MLflow` - `DAGsHub` ### ✅ Model Testing & Validation - `Deepchecks` - `Pytest` ### 🧩 API Development & Deployment - `FastAPI` - `Docker` - `Amazon ECR` - `AWS ECS` - `Elastic Load Balancer (ELB)` ### 📡 Monitoring & Drift Detection - `Arize AI` ### 📊 Visualization & Reporting - `Streamlit` - `GitHub Pages` ### ⚙️ CI/CD Automation - `GitHub Actions` --- ## ✅ Outcomes This pipeline enabled: - ⚡ **Scalable deployment** - 🛠️ **Continuous monitoring** - 📦 **Reproducibility & transparency** - 🔁 **Rapid model iteration** --- ## 📎 Tools & Technologies | Category | Tools Used | |--------------------------|----------------------------------------------------------------------------| | Data Handling | Python, Pandas, NumPy | | Modeling | SVM, Linear Regression, VAR | | MLOps & Tracking | DVC, MLflow, DAGsHub | | Testing & Validation | Deepchecks, Pytest …