Among the phenomena that occur on the surface of the Sun, solar flares may cause several damages, from short circuits in power transmission lines to complete interruptions in telecommunications systems. In order to mitigate these effects, many works have been dedicated to the proposal of mechanisms capable of predicting the occurrence of solar flares. In this context, the present work sought to evaluate two aspects related to machine learning-based solar flare forecasting: (i) the impact of class imbalance in training datasets on the performance of the predictors; and (ii) whether the incorporation of a pre-clustering step prior to the classifiers training contributes to a better prediction.