Machine learning, neural networks and statistical pattern recognition for voltage security: a comparative study