Predicting Job Openings in the AI ​​and Data Market Using Artificial Neural Networks

Abstract

The Artificial Intelligence market exhibits high salary and structural variability across regions and companies. This work proposes the use of Multilayer Perceptron (MLP) artificial neural networks to predict job indicators: salary, work modality, and company size. The model predicted salaries with a mean absolute error of 11.49%, but presented significant limitations in multi-class classifications, with accuracies of 33.6% for work modality (equivalent to random prediction) and 45.8% for company size. The main limitations include: (i) a single architecture not optimized for each task; (ii) lack of class imbalance handling; (iii) lack of feature engineering specific to categorical problems. The results demonstrate the effectiveness of MLPs for salary regression, but indicate the need for specialized models for classification in volatile markets.