Sexism detection is still considered a challenging task because of the linguistic expressions of many online comments. In this study we present a comparative analysis between three supervised learning models: Random Forest, Gradient Boosting and Multilayer Perceptron (MLP). These models were applied to the classification: positive, negative and neutral, in texts with sexist content. The dataset was constructed from one million comments collected from Reddit and Gab platforms, along with 20,000 instances manually labeled by the EDOS team as part of the SemEval 2023 competition. The texts were preprocessed and vectorized using word embeddings that were generated with Word2Vec, thus preserving the semantic relationships between terms. The results obtained indicate that, although the decision tree-based models offered acceptable initial performance, the MLP with 7 hidden layers achieved the best overall performance, outperforming the other models in terms of accuracy, recall and F1-score. Based on the results, the effectiveness of using deep neural networks to capture complex patterns in textual data is shown with this result. Finally, the potential of using more efficient models, such as MLPs, as a viable alternative to more expensive architectures such as transformers, especially in environments with limited computational capacity, is highlighted.