There are many challenges in procedural content generation (PCG), maintaining the quality of content with all the diversity, without any flaws and maintaining the balance in the level of difficulty by dynamic difficulty adjustment (DDA) of the game content in real time based on active inputs are some of the primary issues when concerned with machine generated adaptive content. This subject investigates the efficiency in the use of Procedural content generation by combining the learning based and rule based approach. The research is motivated by the PCG framework based on the learning approach, with the novel approach to address the two primary issues in Super Mario Bros (SMB). To deal with the content quality issue the combination of rule based and learning based techniques is used, which will help to produce the game segments with good quality, also called constructive primitives (CPs) which are all the quality segments in a game level. A DDA algorithm is proposed in the research that will control a CP-based level generator and will adjust the difficulty of levels rapidly based on players' real-time game play experience. The use of model based approach over model free approach will be the main focus of this research where the sampling will be done to minimize content into their respective datasets using the simple random sampling approach without replacement. In a model-free approach, player model is mainly determined by the players' gameplay data, e.g., controllable parameters and playlog in the game, and the feedback from this control parameters, e.g., state changes the affect the "fun" content of the game, while model based approach have the player model which relies on derived values of psychological emotion theories to mainly balance the game difficulty [14].As far as it is known most of the current implementation on SMB game falls under the model free approach and the model based approach is yet to be explored and not many have efficiently implemented the model based approach. The real-time model based approach for Dynamic Difficulty Adjustment (DDA) within the generation of a level is only implemented in the recent work of P. Shi and K. Chen [5], [6] for SMB level adaptation research to date. Learning-Based Procedural Content Generation (LBPCG) framework [15]is the main motivation in our research. The quality evaluation function of the content of the game will be determined from the levels mentioned by the developers. Generating short game segments of good quality will be determined as constructive primitives (CP) [5], [6] and modified Infinite Mario Bros will be used as a test game [3].