This paper presents support vector machines (SVM) as a machine learning technique which is proposed as a method for performing nonlinear equalization in coherent optical CO-OFDM systems. The equalizer can offer a 7.75 dB improvement in optical signal-to-noise ratio (OSNR) compared to Volterra algorithm for 100 Gb/s CO-OFDM signals when considering only nonlinearities to achieve a Bit-Error-Ratio inferior to 10E-3. It is also revealed that SVM can double the transmission distance up to 1600 km over SMF compared to the case of Volterra and Hammerstein. The proposed algorithms are tested in simulation using a coherent 100 Gb/s 4-QAM optical communication system in a legacy optical network setting. We also present a blind-support vector machines (B-SVM) as a machine learning technique which is proposed as a method for performing nonlinear equalization in CO-OFDM systems. The B-SVM equalizer showed more efficiency in combating single-mode fiber (SMF) induced nonlinearities compared to the Volterra. The B-SVM equalizer can reduce the fiber nonlinearity penalty by 1 dB at 2 dBm of launched power improvement compared to Volterra algorithm for 40 Gb/s CO-OFDM signals. It is also shown that SVM-NLE outperforms both V-NLE and full-field DBP-NLE in terms of Q-Factor by ∼6 and 1.5 dB, respectively.