Abstract—Location-based services for various indoor applications are often built upon the results offered by indoor positioning systems. Among many positioning approaches, Wi-Fi received signal strength indicator fingerprinting based techniques are of particular interest because of wide Wi-Fi network deployment in many indoor environments. With the rapid development of computing resources, many deep learning models have been proposed for determining indoor mobile objects showing their superiorities compared to traditional models. However, the hyperparameter tuning procedure for obtaining the most suitable model is very challenging and time-consuming. To relax this circumstance, this paper presents a utilization of Bayesian Optimization to reach the best hyperparameters of a long short-term memory regression model for indoor positioning solutions. In addition, a combination of two well-known dimensionality reduction techniques namely Truncated Singular Value Decomposition and Linear Discriminant Analysis is proposed to enhance the positioning accuracy. The results produced on a public dataset show a considerable improvement of the proposed solution over the others in terms of positioning accuracy, i.e., the mean distance error improved by 3%, 9%, and 24% compared to three state-of-the-art studies.