Aluminium Price Forecasting POC: hybrid statistical and machine learning models for procurement cost optimization.

Project facts & technologies

This block is designed to give analysts, journalists, and AI search systems a discrete, citation-friendly summary of the project. Each row is a clean entity-attribute pair.

Project name

Aluminium Price Forecasting POC

Industry

Manufacturing, Industrial Procurement, Commodity Price Risk

Use case

Real-time aluminium price forecasting for procurement decision support

Engagement type

Proof of Concept (POC)

Statistical models

ARIMA, SARIMA

Machine learning models

XGBoost, Random Forest

Feature engineering

Lags, moving averages, volatility features, macro indicators

Inputs

LME aluminium prices, FX, oil and energy indicators, supply-demand signals, internal procurement history

Deployment

Flask web application with real-time forecast access

Business outcome target

25%+ reduction in aluminium procurement cost

ML outcome target

80%+ accuracy on price-trend predictions

Economic outcome target

20%+ overall company revenue uplift

Why is aluminium procurement so exposed to price volatility?

Aluminium is one of the most strategically important industrial commodities — used in automotive, aerospace, construction, packaging, and consumer products — and its price moves on a complex web of drivers. London Metal Exchange (LME) prices respond to global supply and demand, energy costs, FX shifts, geopolitical events, tariff changes, and inventory positions held at major exchanges. For manufacturers that procure aluminium as a critical raw material, a 5–10% price move in a single month is not unusual, and the margin impact is direct.

In most manufacturing organizations, however, aluminium procurement is still managed reactively. Buyers act on quotes from suppliers, on rough rules of thumb, or on month-end demand pressure — without a structured view of where prices are likely to head over the relevant procurement horizon. Modern hybrid forecasting models, combining classical statistical time-series methods (ARIMA, SARIMA) with machine learning (XGBoost, Random Forest) and rich feature engineering, finally make it economical to convert reactive purchasing into informed, evidence-based procurement.

What problem does the Aluminium Price Forecasting POC solve?

AiSPRY's manufacturing client procures aluminium on an ad-hoc basis without systematically accounting for price volatility — a reactive approach that drives up procurement cost and hurts operational efficiency. The POC was scoped to demonstrate that a forecasting application could change the economics of that decision. Several structural challenges had to be addressed:

Key challenges

• Ad-hoc procurement decisions — purchases made without a structured view of likely near-term price movement, leading to higher average procurement cost.
• Volatile aluminium prices — LME-driven volatility responding to global supply, demand, energy, FX, and geopolitical signals — challenging to forecast with rules of thumb.
• Limited internal forecasting capability — no in-house time-series or machine-learning capability dedicated to commodity price forecasting.
• Multiple, heterogeneous signals — relevant inputs spread across LME data, FX rates, energy prices, supply-demand indicators, and internal procurement history.
• Procurement-team usability — any forecast that requires a data scientist to interpret won't change procurement behaviour day-to-day.
• Risk minimization requirement — any forecasting approach must explicitly quantify and surface market-volatility risk, not just point estimates.

How does the Aluminium Price Forecasting POC work?

The POC is a hybrid forecasting application that ingests aluminium price history alongside macro and supply-demand signals, runs four complementary modeling approaches in parallel, validates each through walk-forward back-testing, and surfaces the best-of-stack forecast through a Flask web app for the procurement team. Confidence bands accompany every forecast so buyers can read both the central prediction and the volatility risk around it.

See Aluminium Price Forecasting in action

A walkthrough of the Flask web app — real-time aluminium price forecasts with confidence bands, buy / hold decision signals, scenario simulation on macro shifts, and the back-tested model selection that underpins the procurement workflow.

Aluminium Price Forecasting POC — hybrid forecasts on demand

Click to play · ARIMA + SARIMA + XGBoost + Random Forest in a Flask web app

• Real-time forecasts — hybrid statistical and ML predictions accessible through a Flask web app
• Confidence-aware output — quantified uncertainty bands, not point estimates
• Buy / hold decision signals — derived from forecast direction and volatility risk flags
• Procurement-ready dashboard — designed for buyers, not data scientists

What is the architecture of the Aluminium Price Forecasting POC?

The POC is built as a five-stage pipeline — from price and market data sources, through ingestion and feature engineering, into a hybrid statistical and machine learning model core, layered with back-testing and risk validation, and surfaced through the Flask web application and procurement decision support layer.

How does the POC minimize risks from market volatility?

The constraint named in the brief — minimize risks from market volatility — is treated as a first-class engineering input rather than a footnote. Volatility-aware forecasting was built into every layer of the POC.

What measurable outcomes does the Aluminium Price Forecasting POC target?

The POC was scoped against three aligned dimensions of success — business, ML, and economic — to validate that hybrid forecasting could justify a production rollout.

Aluminium Price Forecasting POC — frequently asked questions