Mabruroh, I., & Herumurti, D. (2019). Adaptive Non-Playable Character in RPG Game Using Logarithmic Learning for Generalized Classifier Neural Network (L-GCNN). Kinetik: Game Technology, Information System, Computer Network, Computing, Electronics, and Control, 4(2). doi:http://dx.doi.org/10.22219/kinetik.v4i2.755 Receive January 28, 2019; Revise January 29, 2019; Accepted February 02, 2019 KINETIK, Vol. 4, No. 2, May 2019, Pp. 127-136 ISSN : 2503-2259 E-ISSN : 2503-2267 127 Adaptive Non-Playable Character in RPG Game Using Logarithmic Learning for Generalized Classifier Neural Network (L-GCNN) Izza Mabruroh* 1 , Darlis Herumurti 2 1,2 Intsitut Teknologi Sepuluh Nopember Surabaya mabrurohizza@gmail.com *1 , darlis@its-sby.edu 2 Abstract Non-playable Character (NPC) is one of the important characters in the game. An autonomous and adaptive NPC can adjust actions with player actions and environmental conditions. To determine the actions of the NPC, the previous researchers used the Neural Network method but there were weaknesses, namely the action produced was not in accordance with the desired so the accuracy of action was not good. This study overcomes the problem of accuracy of action that is not good in previous studies that use Neural Network to decide on NPC actions by using the Logarithmic Learning method for Generalized Classifier Neural Network (L- GCNN) with 6 input parameters namely NPC health, distance with players, Other NPCs are involved or not, attack power, number of NPCs and NPC levels. In this study, we will discuss the accuracy of L-GCNN in determining the behavior of NPCs so that the NPC gets the optimal decision in attack compared to using other NN methods. While the output is to attack itself, attack in groups and move away. While the output is to attack itself, attack in groups and move away. For testing, this study was tested on RPG games. From the results of the experiments conducted, it shows that the L-GCNN method has better accuracy than the 3 methods compared to 7% better than NN and SVM and 8% better than RBFNN because in the L-GCNN method there is an encapsulation process that is data have the same class will. Whereas the L-GCNN training time is 30% longer than the NN method because on L-GCNN one neuron consists of one data where there are fewer NNs in the hidden layer. So that it has better action accuracy. Keywords: L-GCNN, NPC adaptif, aksi NPC 1. Introduction Since the emergence of the ideas about artificial intelligence, games are one of the items that helped advance AI research [1]. Games not only cause interesting and complex problems for AI to solve, they also provide land for creativity and expression for users [2]. Thus it can be said that the game is a rare domain where there are elements of art and interaction that make the game unique and favorite for AI studies. But not only is AI progressing through research, games have also advanced through AI research [3]. One of the forming elements of the game is Non-Playable Character (NPC) which is becoming an increasingly challenging domain for artificial intelligence techniques because of its complex nature [4]. One method commonly used to regulate NPC behavior is Finite State Machine where this method is a simple method that is easy to implement, predictable in response, flexible and has light computing. But it has the disadvantage of being a condition for a fixed state transition that is not properly used in games because of its predictable nature [5]. Yunifa et al. Applied fuzzy logic and Hierarchical Finite State Machine (HFSM) to regulate the behavior of NPCs in order to emulate human strategies in war games. The results obtained include HFSM successfully modeling the maneuvering behavior of each NPC and the application of fuzzy to manage the behavior of NPCs to successfully outperform NPCs without fuzzy up to 80% [6]. Furthermore, Supeno et al. Conducted a study on close combat games using a fuzzy coordinator where the rule base was used to regulate the fighting action of an NPC. The coordination action of this research is not enough to produce a variety of actions and sometimes the actions are not as desired [7]. Supria et al. Used L-GCNN to improve the accuracy of the introduction of SIBI sign language. In his research proposed the introduction of the Indonesian Language Signal System (SIBI) by using a combination of static features with dynamic features based on Logarithmic