seMLP: Self-evolving Multi-layer Perceptron in Stock Trading Decision Making

Abstract There is a growing interest in automatic crafting of neural network architectures as opposed to expert tuning find the best architecture. On other hand, problem stock trading considered one most dynamic systems that heavily depends on complex trends individual company. This paper proposes novel self-evolving system called Multi-Layer Perceptron (seMLP) which can abstract data and produce an optimum architecture without tuning. seMLP incorporates human cognitive ability concept abstraction into network. Genetic algorithm (GA) used determine capable knowledge data. After determining with minimum width, prunes remove redundant neurons network, thus decreasing density achieving conciseness. evaluated three market sets. The optimized models obtained from are compared benchmarked against state-of-the-art methods. results show automatically choose performing models.