Indoor localization based on WiFi is growing at a fast pace. Among different indoor localization techniques based on WiFi, the fingerprint based indoor localization systems, gained more attention recently. Despite its simplicity, still it needs ample amount of work to build fingerprint database in calibration phase. Furthermore, in online localization phase, for finding a match between observed signal vector and entire fingerprint collection is not feasible. The goal is to minimize both calibration and searching effort in online stage. This paper presents kernel online sequential extreme learning machine (KOS-ELM) based approach to lessen the site survey effort for radio map construction. Furthermore, a hybrid Wireless LAN(WLAN) based approach for online localization phase is presented. Proposed solution incorporate both trilateration and fingerprinting algorithms in localization phase to further improve the localization accuracy. At first, the radio signatures have been collected at few positions from the environment in a usual manner after that based on collected data, the KOS-ELM algorithm has been trained. Later, the trained KOS-ELM was able to successfully predict radio signatures at other unknown coordinates of the environment. In localization stage first, trilateration finds a location based on path loss model, afterward, that location will be fed into K-Nearest Neighbor(KNN) which will search for fingerprint matching within two meter radius of location returned by trilateration algorithm.