BAnalysis-3-2/StockWise-AI

AI-powered inventory optimization system for South African SME retailers. Combines ensemble machine learning (LightGBM, XGBoost, ARIMA and LSTM) with intelligent chatbot for demand forecasting and inventory management. AI Demand Forecasting System Problem Definition Small and medium-sized enterprises (SMEs) in South Africa often struggle with inventory mismanagement due to inaccurate demand forecasts. This project addresses that challenge with an AI-driven solution aligned with the theme of AI Solutions for Industry . The system improves decision-making, reduces waste, and enhances operational efficiency. AI Solution This system forecasts retail demand using strict data thresholds and forensic-grade preprocessing. It routes SKUs to ARIMA or LSTM based on row count and produces transparent, teachable outputs for stakeholder review. It leverages a hybrid AI architecture combining multiple data sources—retail sales, economic indicators, and social media sentiment—with advanced models including LightGBM, XGBoost, ARIMA, and LSTM. The system supports real-time alerts and stakeholder interaction via an integrated chatbot interface. Business Background Inventory mismanagement limits SME scalability and profitability. This project supports South Africa’s 4IR goals by deploying AI-powered forecasting tools to enhance productivity, decision-making, and operational agility. It empowers SMEs to compete more effectively in modern markets. Key Features Modular preprocessing with strict column validation Feature engineering: lag, rolling stats, price change, inventory ratio Model routing: LSTM for 30 or more rows and ARIMA for less than 30 rows Forecast output: SKU-level CSV with model_used and MAE Chatbot interface for forecast lookup and simulation Tools and Techniques Machine Learning Platform: Python with scikit-learn, LightGBM, XGBoost, TensorFlow/Keras Time-Series Forecasting: ARIMA, Long Short-Term Memory (LSTM) Data Processing & Integration: Python libraries for feature extraction, sentiment analysis, and data cleaning Machine Learning Approach Ensemble Modeling: Combines LightGBM, XGBoost, ARIMA, and LSTM for robust forecasts Transfer & Few-Shot Learning: Adapts …