Revisão da demanda regional por vagas de TI no Brasil: uma análise comparativa de modelos de aprendizado de máquina

Abstract

This study develops and compares machine learning models for forecasting regional demand for jobs in the Information Technology (IT) sector in Brazil. Using a dataset of over 66,000 job postings extracted from LinkedIn via web scraping and structured as daily time series, the analysis covers the five Brazilian macro-regions and a specific category for remote jobs. Three machine learning algorithms for forecasting were comparatively evaluated: Support Vector Regression (SVR), feedforward neural networks of the MLP (Multi-Layer Perceptron) type, and Long Short-Term Memory (LSTM) neural networks. The performance of the models, trained individually by region, was measured using the MAE (Mean Absolute Error) and MDA (Mean Directional Accuracy) metrics. The results demonstrate that the feedforward model obtained the best performance, although the SVR model obtained similar results. According to the data, remote job openings, combined with those in the Southeast and South regions, represent more than 90% of the total, highlighting the lower supply in other regions. Therefore, this work demonstrates the viability of the methodology for capturing regional demand patterns, offering a robust quantitative overview of the behavior of the Brazilian IT sector by region and work modality.