📦 Supply Chain Demand Forecasting & Inventory Optimization
Note: The application is currently not hosted on AWS Free Tier to avoid unnecessary charges. It will be redeployed later and extended to support forecasting on any user-uploaded dataset.
An AI-powered application that forecasts product demand and generates intelligent inventory restocking recommendations. Built using Facebook Prophet, custom hybrid clustering algorithms, and deployed on AWS EC2 using Streamlit.
- 📈 Time Series Forecasting (30-day horizon) using Facebook Prophet Library
- 🧠 Custom Hybrid Clustering to segment products by behavior
- 🔁 Inventory Action Recommendations based on predicted vs. current stock
- 📉 Confidence Intervals for better decision-making
- 🖥️ Deployed on an EC2 Ubuntu instance using Python + Streamlit
Firstly, the user access the site using web browser. The Service used for running the application is AWS Elastic Cloud Compute (EC-2) and the deploying is done on streamlit At the backed of the app, there exists a forecasting engine designed using the Prophet library by Meta and there also exists a clustering model which is a hybrid of K- Means clustering, DBSCAN and PCA
Why K means ? To segment products into k distinct clusters based on their features like quantity sold, frequency of sales, and demand trends. Its advantanges- Fast and effective on large data sets' analysis, well separared and spherical clusters, helps in identifying high/low volume of products
Why DBSCAN ? To detect outliers and irregular demand patterns. Unlike K-Means, DBSCAN doesn’t assume the number of clusters. Identifying anomalous or erratic-selling products and filter out the products that don't follow the normal trend
Why PCA ? To reduce the number of features while preserving the important patterns. Dataset may include high-dimensional vectors (like sales over time), so PCA reduces that to 2 principal components for clustering. PCA simplifies the problem, allowing clustering to focus on core patterns rather than noise.