This thesis presents the development of a software module built in Python using the FastAPI framework, designed to perform sales forecasting based on a dataset of historical invoices. The system employs the XGBoost machine learning model to generate accurate predictions, chosen for its high performance and robustness in handling structured tabular data. To ensure transparency and interpretability, the module integrates the SHAP (SHapley Additive exPlanations) library, which allows the analysis of individual feature contributions to model decisions, in accordance with the principles of Explainable Artificial Intelligence (XAI). The results and explanations are then translated into natural language and stored in a ChromaDB vector database, enabling interaction with a Large Language Model (LLM) capable of answering questions about the predictions and the reasoning behind them. The project combines techniques from machine learning, XAI, and natural language processing to create a predictive system that is transparent, interactive, and easily applicable in business contexts.
L'elaborato descrive lo sviluppo di un modulo software realizzato in Python utilizzando il framework FastAPI, progettato per effettuare previsioni di vendita a partire da un dataset di fatture storiche. Il sistema sfrutta il modello di machine learning XGBoost per la generazione delle previsioni, selezionato per le sue elevate prestazioni e la capacità di gestire dati tabellari complessi. Per garantire trasparenza e interpretabilità, il modulo integra la libreria SHAP (SHapley Additive exPlanations), che consente di analizzare l’impatto delle singole variabili sulle decisioni del modello, in linea con i principi della Explainable Artificial Intelligence (XAI). I risultati e le spiegazioni prodotte vengono tradotti in linguaggio naturale e archiviati in un database vettoriale ChromaDB, rendendo possibile l’interazione con un Large Language Model (LLM) in grado di rispondere a domande sulle previsioni e sulle motivazioni che le supportano. Il progetto unisce quindi tecniche di machine learning, XAI e natural language processing, con l’obiettivo di creare uno strumento predittivo trasparente, interattivo e facilmente integrabile in contesti aziendali.
Progettazione di un Motore di Previsione Vendite per Sistemi ERP Aziendali
STIGLET, OLIVER FLORIN
2024/2025
Abstract
This thesis presents the development of a software module built in Python using the FastAPI framework, designed to perform sales forecasting based on a dataset of historical invoices. The system employs the XGBoost machine learning model to generate accurate predictions, chosen for its high performance and robustness in handling structured tabular data. To ensure transparency and interpretability, the module integrates the SHAP (SHapley Additive exPlanations) library, which allows the analysis of individual feature contributions to model decisions, in accordance with the principles of Explainable Artificial Intelligence (XAI). The results and explanations are then translated into natural language and stored in a ChromaDB vector database, enabling interaction with a Large Language Model (LLM) capable of answering questions about the predictions and the reasoning behind them. The project combines techniques from machine learning, XAI, and natural language processing to create a predictive system that is transparent, interactive, and easily applicable in business contexts.| File | Dimensione | Formato | |
|---|---|---|---|
|
thesis.pdf
accesso aperto
Dimensione
17.33 MB
Formato
Adobe PDF
|
17.33 MB | Adobe PDF | Visualizza/Apri |
The text of this website © Università degli studi di Padova. Full Text are published under a non-exclusive license. Metadata are under a CC0 License
https://hdl.handle.net/20.500.12608/102074