Machine Learning Techniques: A Tool for Learning and Planning
Babatunde Akinbode · International Conference on Computing Technology and Information Management · 2014
To invest in natural environments, an investor must decide the best action to take accord ing to its current situation and goal, a problem that c an be represented as a Markov Decision Process (MDP). In general, it is assumed that a reasonable state representation and transition model can be provided by the user to the system. When dealing with complex domains, however, it is not always easy or possible to provi de such information. In this paper, a system is descri bed that can automatically produce a state abstraction and can learn a transition function over such abstracte d states, called qualitative states. A qualitative state is a group of states with similar properties and rewards. They are induced from the reward function using decision trees. The transition model represented as a MDP is learned using a Bayesian network learning algorithm. The outcome of this combined learning process produces a very compact MDP that can be efficiently solved using standard techniques. We show experimentally that this approach can learn efficiently a reasonable policy that an investor ca n invest upon in large and complex domains.