https://repositorio.ufu.br/handle/123456789/17752 2026-09-06T06:50:17Z https://repositorio.ufu.br/handle/123456789/50109 Title: Efeito de diferentes dosagens do adubo orgânico à base de casca de maracujá na cultura da mandioca no Cerrado 2026-03-04T00:00:00Z https://repositorio.ufu.br/handle/123456789/49881 Title: Modelo preditivo do ciclo de colheita de sementes de milho 2026-08-05T00:00:00Z https://repositorio.ufu.br/handle/123456789/49779 Title: Impacto da COVID-19 na saúde cardiovascular 2026-08-07T00:00:00Z https://repositorio.ufu.br/handle/123456789/49034 Title: Entre o acaso e a inteligência: uma comparação entre investimentos na B3 baseados em aprendizado de máquina e o jogo da roleta Abstract: This study aims to investigate the use of machine learning algorithms as support for medium-term investment strategies on the Brazilian Stock Exchange (B3), comparing their percentage performance with that of casino roulette betting systems. In this context, three supervised classification models are tested, namely: k-nearest neighbors (k-NN), decision tree, and random forest. Thus, the signals generated by these models are employed in buy-only transactions of the target assets — without any sell orders — allowing evaluation of their wealth multiplication potential from the perspective of the risk-return trade-off inherent in the equity market. Regarding the assets analyzed, over 30 stocks comprising the Ibovespa index are initially examined; at the end of this selection process, four of them — assessed on weekly and monthly price scales — are defined as the objects of study: VALE3, ITUB4, PETR4, and SBSP3. In parallel with the computational modeling related to the stock market, a computational system based on the probabilistic logic of casino roulette betting is implemented to establish a benchmark of random nature for comparative purposes. Within this system, the Streets of Gold betting strategy is applied in pursuit of profit maximization and mitigation of the randomness effects inherent to the game. The results indicate that, in absolute terms, no model demonstrates superiority across all assets and time horizons; however, certain combinations achieve significantly higher returns than the betting system under identical conditions of time (rounds) and available capital, delivering not only substantial financial gains but also lower exposure to wealth risk. Thus, this study contributes to the existing academic literature by demonstrating the potential of machine learning in formulating financial strategies that provide practical support to investors seeking quantitative alternatives in highly volatile market scenarios. Additionally, it underscores the purely recreational nature of betting systems. 2025-11-10T00:00:00Z