Handling imbalanced datasets through Optimum-Path Forest

In the last decade, machine learning-based approaches became capable of performing a wide range complex tasks sometimes better than humans, demanding fraction time. Such an advance is partially due to exponential growth in amount data available, which makes it possible extract trustworthy real-world information from them. However, such generally imbalanced since some phenomena are more likely others. behavior yields considerable influence on learning model's performance becomes biased frequent receives. Despite methods, graph-based approach has attracted notoriety outstanding over many applications, i.e., Optimum-Path Forest (OPF). this paper, we propose three OPF-based strategies deal with imbalance problem: $\text{O}^2$PF and OPF-US, novel for oversampling undersampling, respectively, as well hybrid strategy combining both approaches. The paper also introduces set variants concerning mentioned above. Results compared against several state-of-the-art techniques public private datasets confirm robustness proposed