This thesis investigates a new approach to enhancing random forests by integrating principles of natural selection inspired by genetic algorithms. Traditional decision tree induction methods often rely on specific heuristics, potentially introducing biases and limiting solution space exploration. While random forests mitigate some of these issues by averaging predictions from multiple trees, there is still room for improvement in tree construction methods for finding optimal forest configurations. This work proposes a new random forest induction method, utilizing genetic algorithms to evolve tree structures dynamically. The evolutionary approach aims to overcome the limitations of traditional methods by fostering more diverse and adaptable models. The research evaluates the performance of this evolutionary method in practical classification tasks, comparing it against the established Scikit-learn random forest implementation. Results indicate that while Scikit-learn's random forest remains the preferred choice for efficiency and accuracy, the evolutionary approach presents a viable alternative for applications emphasizing interpretability and robustness.