Combining Machine Learning with Seasonal-Trend Decomposition using LOESS in Power BI

Aug 3, 2026

Time series analysis supports forecasting and trend discovery across industries by revealing temporal dependence, seasonality, noise, and irregular deviations. Seasonal-Trend decomposition using LOESS (STL) clarifies the seasonal-trend structure, enabling better interpretation of residual patterns when paired with machine learning. A hybrid workflow integrates STL with a Random Forest regressor inside Power BI dashboards, combining decomposition-driven features with predictive modeling. Results indicate improved robustness and enables sophisticated interactive forecasts directly in business intelligence environments.

LanguageEnglish

CategoryResearch & Report

FormatPDF

Pages9

Reading Time23 min

AccessFree to download

Introduction

Seasonal-Trend Decomposition using LOESS

STL components: trend, seasonality, residuals

LOESS-based iterative decomposition

Combining Machine Learning with Seasonal-Trend Decomposition using LOESS in Power BI Assoc. Prof. Dr. Yanka Aleksandrova University of Economics – Varna, Varna, Bulgaria [yalexandrova@ue-varna.bg](yalexandrova@ue-varna.bg) Chief Assist. Prof. Dr. Mihail Radev University of Economics – Varna, Varna, Bulgaria [radev@ue-varna.bg](radev@ue-varna.bg) Abstract Time series analysis has been extensively used for forecasting in various industries. A method frequently used for decomposition of time series is Seasonal-Trend decomposition using LOESS (STL). In combination with the machine learning approaches, STL is a helpful method to analyze the seasonal-trend structure of complicated time series. This hybrid approach helps interpret seasonality, trends, and other residual patterns better than when using only predictive machine learning models. The explanation and interpretation of the models can be effectively implemented in the context of Business Intelligence and analytical platforms. In the current paper, a practical approach involving the integration of STL with Random Forest regressor in Power BI has been proposed. It is evidenced from the results that integration of STL decomposition with machine learning provides a robust analytical tool and this allows user to perform a sophisticated time series forecasts right within the engaged interactive dashboards. Keywords: seasonal-trend decomposition, STL, machine learning, random forest, forecasting JEL Code: C530 DOI: 10.56065/IJUSV-ESS/2024.13.1.81 Introduction In recent years, time series analysis has become a crucial tool for forecasting and trend analysis in various industries. Time series analysis is focused on exploring data collected over time to uncover trends, patterns and relationships. The identified patterns and trends can be used for forecasting, modeling relationships between multiple time series, detection of anomalies, etc. Time series analysis addresses the key characteristics of temporal dependent data. One of the major characteristics is temporal dependence, meaning that the observations are not independent but depend on their prior values which results in specific temporal patterns. Seasonality is another important aspect, as often data fluctuate at fixed intervals. Data may also show trends, indicating along-term upward or downward movement over time. At the same time, time series data often contain noise or irregularities, which are random fluctuations or deviations that do not follow predictable patterns. Various methods for time series analysis can be used depending on the analysis objectives. The most popular and widely used methods can be summarized into the following groups (Box et al., 2016),(Huang and Petukhina, 2022),(Mahmud et al., 2023): ● Methods for decomposition of the time series data into three main components: trend, seasonality and noise ● Smoothing methods to remove noise and identify long-term patterns and trends ● Autoregressive methods to model the dependency of current from previous observations ● Regression-based methods to capture trends and seasonality ● Clustering and classification to group and categorize time series, etc. Alongside the classic statistical methods for time series analysis, various machine learning and hybrid models can be implemented for advanced predictions and patterns recognition, thus leveraging the potential of AI algorithms(Armyanova, 2022) . Some of the popular algorithms for machine learning time series analysis include Long Short-Term Memory, Transformer Models, hybrid models combining statistical methods like ARIMA with machine learning algorithms like Neural Networks, Random Forest, XGBoost among others. A hybrid approach for combining classic well-known and established statistical methods with advanced machine learning algorithms can significantly improve the predictability and interpretability of the time series analytical models thus allowing for the identifica

FAQs

What does STL decompose in a time series?

STL decomposes a time series into three components: trend, seasonality, and residual (remainder) that captures random unpredictable changes.

Why combine STL with machine learning models?

The hybrid approach improves interpretability by separating seasonal and trend structure and then learning patterns using machine learning, producing more robust forecasts than using predictive models alone.

How is the proposed method implemented in Power BI?

The paper proposes integrating STL decomposition with a Random Forest regressor within Power BI so users can perform advanced time series forecasting inside interactive dashboards.

Aug 3, 2026

Upload by Evangeline