In spite of the amount of research done in the prediction of the progression of mild cognitive impaired (MCI) subjects to Alzheimer's disease (AD), there is still room for further improvement. Sophisticated methods have been proposed, some reaching classification accuracies of up to 85%. In the present paper, we propose a combination of simple ideas to determine if they allow to obtain similar accuracies when predicting MCI to AD conversion. We present three approaches making use of ADNI database. We set a performance baseline using only demographic and clinical data (gender, education level, APOE4, MMSE, CDR sum of boxes, ADASCog) that provides a balanced accuracy of 76% (AUC of 0.84). When using imaging data, an important finding is that when an SVM is trained for discriminating between cognitive normal (CN) subjects and AD patients, and the resulting classifier is applied to MCI subjects to predict conversion, performance using FDG PET data improves to 76% of balanced accuracy and an AUC of 0.82. The third approach, consisting of multimodal data, namely the combination of the scores obtained from SVM for T1w and FDG PET data, and the demographic and clinical data, provided the best prediction results (80% balanced accuracy, AUC of 0.88). These prediction accuracies, resulting from the combination simple ideas, are in line with state-of-the-art results, and provide a new baseline to compare more sophisticated methods against. All the code of the framework and the experiments will be publicly available at https://gitlab.icm-institute.org/aramislab/AD-ML.