Projects

Tanziforecast

Production Forecasting with Multi-Level Daily Predictions to Support Data-Driven Planning

Funding: IFAB for PMI

Enabling technology: Machine Learning

Sustainable Development Goals

The Tanziforecast project involved the development of an advanced web application for forecasting daily production volumes, designed to support operational planning in a highly variable industrial environment.

The platform uses an intelligent production forecasting system based on advanced machine learning algorithms, including LightGBM, trained on historical daily production data. The system generates multi-level forecasts across four aggregation levels: total production, customer, product type and packaging type. Built in Streamlit and accessible through role-based authentication, the solution supports dynamic data uploads, automatic model updates and report generation to inform operational decision-making.

Objectives

The project was created to provide the company with a structured tool for forecasting daily production volumes. Its purpose is to support data-driven operational planning, improve the allocation of production resources, reduce uncertainty in workload management and enable forecasting analyses at different levels of detail, including customer, product line and packaging type.

Initial Challenge

The main challenge was to develop a forecasting system capable of managing the complexity and heterogeneity of production data. These data were characterised by fluctuating customer demand, structural differences across product lines and packaging types, the need to perform analyses at multiple aggregation levels and the requirement to integrate the solution into existing operational workflows.

Solution

The solution uses advanced machine learning algorithms for time-series forecasting, including models such as LightGBM, combined with hierarchical structures for multi-level predictions. It also segments production by customer, category and component, ensuring consistency across the different forecasting levels. Specific methods are applied to irregular demand patterns to improve the accuracy of forecasts for intermittent-demand products.

The system provides:

daily forecasts of production volumes;
multi-level analyses across four aggregation levels;
dynamic data updates through a dedicated upload function;
automatic report generation to support management decisions.

Developed as an interactive web application, the platform enables company users to explore forecasts through dynamic dashboards and advanced data visualisation tools. This makes it easier to interpret the results and compare different forecasting scenarios. The entire system has been designed to be scalable, updatable and compatible with future developments in the company’s data infrastructure.

Benefits

The introduction of the platform enabled the company to structure its production planning process around quantitative data, supporting a predictive approach to operational management.

The main benefits include:

greater accuracy in production volume planning;
optimised use of production lines;
improved management of demand peaks and fluctuations;
reduced margins of error in operational estimates;
a stronger data-driven culture across the company.

IFAB’s Role

IFAB managed the project’s entire technical development process, drawing on its expertise in data engineering, statistical modelling and deep learning. The initial phase focused on collecting and preparing historical sales and production data. This included data cleaning, the management of missing values and outliers, and time-series-specific feature engineering. These activities required an in-depth understanding of the structure of the company’s production data.

Based on this work, several forecasting models were developed, trained and compared. These ranged from established statistical methods, such as ARIMA and exponential smoothing, to machine learning and deep learning models, including gradient boosting and recurrent neural networks such as LSTM and GRU. Particular attention was given to identifying seasonal patterns and recurring demand peaks, such as those associated with public holidays. These are especially relevant in the cured meat market and require targeted adjustments to production planning. The selected model was ultimately integrated into the company’s production workflow and placed under continuous monitoring, with regular updates designed to adapt it to changing market dynamics.

Participation in Tanziforecast enabled Fondazione IFAB to strengthen a broad set of highly relevant applied capabilities. These include the management and preparation of complex industrial data, the selection and validation of time-series forecasting architectures, and the integration of predictive models into real production environments. This experience further demonstrates IFAB’s ability to translate advanced artificial intelligence tools into practical operational solutions, including for medium-sized manufacturing companies within the Emilia-Romagna industrial ecosystem.