Depression is a mental disorder that causes self-mutilation or suicide. Repetitive transcranial magnetic stimulation (rTMS) is a treatment way for major depressive disorder (MDD). This paper proposes a novel method to predict the effectiveness of rTMS for depressed patients using electroencephalogram (EEG) signals. In this paper, linear and nonlinear features of multi-dimensional EEG signals are extracted and fed into different neural network (NN) architectures and machine learning (ML) techniques. The cascade forward neural network (CFNN) outperforms other machine learning techniques. The best CFNN architecture consists of 11 layers where each layer has ten neurons. The results show that CFNN achieves promising classification accuracy of 97.10%. The reported results for the CFNN architecture are reliable and robust because 10-fold cross-validation strategy is used to avoid bias. The designed CFNN architecture is simple, and has fewer number of layer than pretrained architectures. The proposed model can be used in clinics and hospitals as a computer-assisted prediction system for predicting the outcome of rTMS treatment for MDD patients.