Forecasting the Absolute Volume of Job Search Activities using Auto-regressive Neural Networks and Extreme Learning Machines Combined

The identification of the business cycle turn points is a challenging task. In this context, the evolution of new data typologies as internet searches data strongly impacted the construction of approaches practical to forecast macroeconomic variables. This work emphasizes the significance of predicting turning points in the business cycle and the potential of Google Trends data in identifying inflection points in data series in labor markets. In this context, the aggregate search behavior analysis could be very relevant to be included and integrated into approaches practical to anticipate turning points in the business cycle. The study aims to examine labor market searches to identify data anomalies that can help anticipate changes in relevant macroeconomic variables. Several forecasting approaches are investigated to predict interest in the labor market over time. Combining forecasts of the different machine learning approaches is beneficial due to the inherent uncertainty associated with each method, and simple averaging may be more accurate than a complex method based on correlations between forecast errors. In this work, we have considered two machine learning approaches, the first one based on autoregressive neural networks and a second one as an ELM extreme learning helpful machine to predict the search behavior better. The findings suggest that combined forecasting approaches are more effective than the classic auto-ARIMA method and can identify innovations in data structures and forecast future changes in the labor market's macroeconomic variables