Today, large volumes of complex data are collected in many application domains such as health, finance and business. However, using traditional data visualization techniques, it is challenging to visualize abstract information to gain valuable insights into complex multidimensional datasets. One major challenge is the higher cognitive load in interpreting information. In this context, 3D metaphor-based information visualization has become a key research area in helping to gain useful insight into abstract data. Therefore, it has become critical to investigate the evolution of 3D metaphors with HCI techniques to minimize the cognitive load on the human brain. However, there are only a few recent reviews can be found for 3D metaphor-based data visualization. Therefore, this paper provides a comprehensive review of multidimensional data visualization by investigating the evolution of 3D metaphoric data visualization and interaction techniques to minimize the cognitive load on the human brain. Complying with PRISMA (Preferred Reporting Items for Systematic Reviews and Meta-Analyses) guidelines this paper performs a systematic review of 3D metaphor-based data visualizations. This paper contributes to advancing the present state of knowledge in 3D metaphoric data visualization by critically analyzing the evolution of interactive 3D metaphors for information visualization. Further, this review identifies six main 3D metaphor categories and ten cognitive load minimizing techniques used in modern data visualization. In addition, this paper contributes three taxonomies by synthesizing the literature with a critical review of the strengths and weaknesses of metaphors. Finally, the paper discusses potential exploration paths for future research improvements.

In the recent development of the Internet of Things (IoT), Artificial Intelligence (AI) plays a significant role in enabling cognitive IoT applications. Among popular IoT applications, location-based services are considered one of the primary applications where the real-time location of a moving object is estimated. In recent works, AI-based techniques have been investigated to the indoor localization problem, showing significant advantages over deterministic and probabilistic algorithms used for indoor localization. This paper presents a feasibility study of using Bluetooth Low Energy (BLE) and Feed Forward Neural Networks (FFNN) for indoor localization applications. The signal strength values received from thirteen different BLE ibeacon nodes placed in an indoor environment were trained using a Feed-Forward Neural Network (FFNN). The FFNN was tested under other hyper-parameter conditions. The prediction model provides reasonably good accuracy in classifying the correct zone of 86% when batch size is 100 under the learning rate of 0.01.Hence the FFNN could be used to implement on location-based IoT applications.

Learning-based localization plays a significant role in wireless indoor localization problems over deterministic or probabilistic-based methods. Recent works on machine learning-based indoor localization show the high accuracy of predicting over traditional localization methods existing. This paper presents a Received Signal Strength (RSS) based improved localization method called TreeLoc(Tree-Based Localization). This novel method is based on ensemble learning trees. Popular Decision Tree Regressor (DTR), Random Forest Regression (RFR), and Extra Tree Regressor have been investigated to develop the novel TreeLoc method. Out of the tested algorithm, the TreeLoc algorithm showed better performances in position estimation for indoor environments with RMSE 8.79 for the x coordinate and 8.83 for the y coordinate.