Although different approaches have been developed for indoor localization, methods using the signal map called "fingerprint" of the space keep their popularity. Most important advantages of these methods are; not requiring extra cost and using existing established Wireless Access Points (WAP). In these systems, localization consists of two steps: (i) generating a signal map and (ii) locating the person over the map. The number and position of WAPs used to generate the signal map are important factors affecting the cost of the system, localization accuracy and operating speed. This work proposes a minimization method that avoids ineffective KENs to create a lower cost and faster system without compromising localization accuracy. For this purpose, Principal Component Analysis (PCA) method which is a successful dimension reduction tool has been utilized. The proposed system consists of a reduced signal map and a localization model using three different machine learning methods (K-Nearest Neighbor, Support Vector Machines, Linear Discriminant Analysis) chosen by the user. As a result of the application, it was seen that the model with 70% less WAPs was able to detect the position of the person with 91% accuracy using K-Nearest Neighbors method and 75% (over the test time) faster compared to the model of non-reduced signal map.