Neural networks have already proved capable of handling noisy and unstructured data such as hand-written texts, images, sounds, and real-world object classification based on an incomplete description. There has been some focus on heuristic learning in which a heuristic function from the workouts using a machine learning model can be automatically induced. Two types of heuristics algorithm are utilized, firstly, an admissible heuristic or informed search is utilized to assess the expense of achieving the objective state in the heuristic algorithm. In this research, the objective state would be for the Rubik’s Cube to be in a solved state. All together for a heuristic to be permissible to heuristic, the assessed cost should dependably be lower than or equivalent to the actual cost of achieving the objective state. The heuristic algorithm utilizes the admissible heuristic to discover an expected ideal way to the objective state from the current state. While an inadmissible heuristic means that it might sometimes find the non-optimal path. By randomly the estimate, it can choose between path cost and search cost. Rubik's Cube makes use of mathematical group theory, which has helped deduce specific algorithms. Furthermore, the fact that there are distinct subgroups in the Rubik Cube group enables the puzzle to be learned and mastered by moving through different "difficulty levels" in itself. Implementing neural network approaches to solve the Rubik's Cube have struggled to succeed without human help and have had to rely on hand-engineered features and group theory to systematically find solutions. The finding of this research paper is to automatically obtain valuable heuristic functions so that the entire problem-solving process is free of human knowledge. This also helps to use heuristic
algorithm rules that perform random walks back from a goal state and try to learn how far they have reached the goal. Several tests ran to compare the different heuristics. Firstly, the comparison is done between the regular heuristics (human-made heuristic, the trained model using heuristic rules), with each other. Then, the comparison is done between the learned heuristics (the learned model from regular heuristic), with each other. Finally, compared the best heuristic from the first group to the respective one from the second group. The results are analyzed as the performance of different heuristics is reasonable. This is due to the fact that the number of nodes are expanded to include A^* algorithm.
Keywords: Neural Network, Rubik’s cube, Heuristic learning, Admissible Heuristics, Inadmissible.