Dyslexia or reading impairment is a reading disorder that can have an effect on both children and adults. Though reading impairment affects the reading abilities of an individual it does not have an effect on the intelligence quotient of the individual. It has been proven that eye-gaze behavior can be used as a good indicator of the reading abilities of a person. Therefore, using eye-gaze behavior and the eye-gaze patterns as a determinant of a good reader and a poor reader can help with the identification of reading impairment. An early detection of reading impairment can help the person in getting the required help as soon as possible. In this work, the detection of reading impairment has been modeled as a reinforcement learning problem using a model-based learning approach. The proposed model is named as RIDynaQ which stands for Reading Impairment detection using DynaQ wherein the eye-gaze behavior of both dyslexic and non-dyslexic children was considered and the optimal gaze targets and the optimal policy for reading for both good and poor readers were obtained. Also, a comparison was drawn amidst the pre-existing model-free approach and model-based approach. The main difference between the existing Q learning based approach and RIDynaQ is the use of simulated experiences and developing a model of the environment which does not exist in Q Learning based approach but is used in the RIDynaQ model.