Sales Forecasting & Inventory Planning with Machine Learning

A system that forecasts daily sales for the next 14 days and recommends how much stock to reorder, so the business never runs out or overstocks.

How it works:

1. Prepared 900 days of sales data and built time-based features (day of week, previous days' and weeks' sales).

2. Compared two models (Ridge and Gradient Boosting) using proper time-series cross-validation and picked the best one.

3. Benchmarked against the traditional method of repeating last week's sales.

4. Calculated expected demand during the supplier's lead time and suggested an order quantity based on current stock.

5. Web interface with a forecast chart and table, plus a REST API.

Result: 5.3% forecast error vs 7.8% for the traditional method (about 32% lower error), with 13 passing automated tests. Built on sample sales data.