By growing the demand for location based services in indoor environments in recent years, fingerprint based indoor localization has attracted many researchers' interest. The fingerprint localization method works based on received signal strength (RSS) in Wireless Sensor Networks (WSNs). This methods uses RSS measurements from available transmitter sensors, which are collected by a smart phone with internal sensors. In this paper, we propose a novel algorithm that takes the advantages of deep learning, extreme learning machine (ELM) and high level extracted features by autoencoder, to improve the localization performance in the feature extraction and the classification. Furthermore, as the fingerprint database needs to be updated (due to the dynamic nature of environment), we also increase the number of training data, in order to improve the localization performance, gradually. Simulation results indicate that the proposed method provides a significant improvement in localization performance, by using high level extracted features by autoencoder, and increasing the number of training data.