Backorder Prediction for the Defense Logistics Agency
My deep-learning contribution to a Defense Logistics Agency study: an LSTM sequence-to-sequence model in TensorFlow forecasting 12-month backorders across ~30K SKUs.
The full report covers a broader study across classical ML and neural-network approaches to demand and backorder forecasting for the Defense Logistics Agency. My work focused on the deep-learning track: an LSTM sequence-to-sequence model in TensorFlow, trained on five years of demand history across roughly 30K SKUs to predict 12-month backorders.
The framing mattered more than the headline numbers. A backorder you can flag a quarter ahead is a procurement decision (re-buy, expedite, qualify a second supplier) instead of an after-the-fact stockout. The full technical report is publicly available through DTIC.