A B S T R A C T
This study introduces a novel machine learning-based methodology for detecting and forecasting the strength of weak signals in the labor market, using Greece as a case study and utilizing Eurostat time series data (2000–2023). Weak signals, conceptualized as subtle anomalies within otherwise stable labor market indicators, were identified through the Isolation Forest algorithm and projected using a Long Short-Term Memory neural network model. Findings highlight structural instability in male manufacturing employment and wholesale/ retail trade, contrasted by stable trends in sectors such as agriculture, education, and public administration. This study contributes to labor market foresight by integrating anomaly detection with predictive analytics, offering valuable insights for proactive, scenario-based policy design in support of a sustainable and adaptive future of work.
Introduction
The concept of weak signals was originally introduced by Ansoff (1975) to refer to subtle emerging indicators that are too incomplete to allow accurate assessment of their impact. In complex systems, weak signals may refer to technical anomalies or performance variabilities (Van Veen & Ortt, 2021; Yu et al., 2022). When these signals are combined with systemic noise, they can produce detectable effects on the evolutionary capabilities of the system in question (Yu et al., 2022). Their timely detection and interpretation have strategic value, as they may signify impending change and the need for anticipatory action (Ha et al., 2023).
In this study, we propose a machine learning approach to identify potential weak signals as variations in performance in otherwise “strong” signals within the labor market, such as employment distribution per sector, economic performance indices, and rates of participation in training. The analysis data are obtained from the labor market-related time series from Eurostat from 2000 to 2023, focusing on Greece.
The detected anomalies (weak signals) are projected into the future to evaluate their significance and are compared to existing forecasts for the labor market. To this end, we utilized the isolation forest algorithm, which is ideal for detecting outliers and anomalies in data (Xu et al., 2023). The algorithm analyzes the variations in the variables and identifies those that show unusual changes, possibly due to impending changes in the labor market. To predict the development and strength of the detected weak signals, the Long Short-Term Memory (LSTM) neural network model was applied, as it is the most suitable for time-series processing and identifying future development trends (Chen et al., 2023).
The forecast analysis (2024–2030) revealed that while overall male and female employment in Greece was expected to follow stable trajectories, weak-signal intensity was strongest in manufacturing for males and wholesale/retail trade (both genders), suggesting that these sectors might have undergone significant structural shifts in contrast to more stable sectors such as agriculture, education, public administration and defense, and accommodation.
Given the limited research on combined labor market foresight and forecasting techniques (Kanzola & Petrakis, 2024), this study aims to analyze and detect important labor market anomaly indicators that could produce significant insights for shaping the future of work and employment. For instance, employment pattern variations in specific economic sectors could indicate the need for targeted active labor market policies to influence economic activity in those sectors. Pairing this approach with scenario analysis on proactive labor market policies, this study is relevant for policymakers and governments focusing on creating an early warning system for impending structural transformations. This, in turn, enables proactive, rather than reactive, policymaking in areas such as education and training, labor activation, and social and fertility policies, which are cardinal for mobilizing labor force participation and improving the interpretation of certain economic variables such as unemployment.
Για να διαβάσετε το υπόλοιπο άρθρο, πατήστε εδώ.
