PREDICTING UNEMPLOYMENT RATE FLUCTUATIONS USING WEB SEARCH INDEX: AN LSTM-BASED APPROACH

Abstract

The accurate prediction of unemployment rates is critical for economic planning and policy-making. Traditional forecasting models often rely on historical economic indicators, which may exhibit lags and limited timeliness. This study explores the potential of incorporating web search query data as a real-time predictor to enhance the accuracy of unemployment rate forecasts. Utilizing an Long Short-Term Memory (LSTM) neural network, the model analyzes time-series data of Google Trends search indices for job-related terms alongside historical unemployment data. The research focuses on evaluating the predictive performance of the LSTM approach compared to conventional autoregressive models. Results indicate that the LSTM model, enriched with web search data, achieves superior forecasting accuracy, capturing non-linear patterns and short-term fluctuations more effectively. The findings underscore the value of big data, such as web search indices, in complementing traditional economic metrics for real-time socioeconomic forecasting. This approach offers policymakers and researchers a timely tool for anticipating labor market dynamics.