SME Sales Forecasting and Drift Detection Using Neural Networks and Public Macroeconomic Indicators

Shinn, Kyaw May Pyone (2025) SME Sales Forecasting and Drift Detection Using Neural Networks and Public Macroeconomic Indicators. Masters thesis, Dublin, National College of Ireland.

Preview	PDF (Master of Science) Download (835kB) | Preview
Preview	PDF (Configuration Manual) Download (960kB) | Preview

Abstract

Small and medium-sized enterprises (SMEs) are highly exposed to macroeconomic volatility but often lack advanced analytics to anticipate adverse performance trends. This research investigates whether combining time-series forecasting with drift detection can provide early warning signals for SME sales instability. Public macroeconomic indicators from Alpha Vantage and FRED were integrated with a synthetically generated panel of 50 SME-like firms observed daily from 2021 to 2023. After extensive feature engineering, a multiple linear regression (MLR) model and a Long Short-Term Memory (LSTM) network were trained to forecast next-day sales, while a feedforward autoencoder was used to detect deviations from normal operational patterns. The Multiple Linear Regression model achieved an R² of 0.8547 and an RMSE of 180.1445 on the held-out test period. The LSTM achieved an R² of 0.8322 and an RMSE of 198.9795 on the same test period. Compared with the baseline regression model, the LSTM’s RMSE was 10.45% higher, and its R² was 2.26% lower, indicating that the deep learning model did not provide a measurable performance gain under SME-style data constraints. The autoencoder attained high recall (0.60) at the cost of low precision (0.02), making it suitable as a sensitive drift indicator. The study contributes a reproducible pipeline that integrates macroeconomic data, forecasting models and anomaly detection into a business intelligence-oriented framework, and provides evidence that well-engineered linear models can match or exceed deep learning performance in SME-scale settings.

Item Type:	Thesis (Masters)
Supervisors:	Name Email Muntean, Cristina Hava UNSPECIFIED
Uncontrolled Keywords:	Business Intelligence; LSTM; Autoencoder; Anomaly Detection; Time-Series Forecasting
Subjects:	Q Science > QA Mathematics > Electronic computers. Computer science T Technology > T Technology (General) > Information Technology > Electronic computers. Computer science H Social Sciences > Economics > Business H Social Sciences > HD Industries. Land use. Labor > Small Business Sector
Divisions:	School of Computing > Master of Science in Data Analytics
Depositing User:	Ciara O'Brien
Date Deposited:	09 Sep 2026 09:06
Last Modified:	09 Sep 2026 09:06
URI:	https://norma.ncirl.ie/id/eprint/9911

Actions (login required)

View Item