Estimating Object Location in RF Communication by Using RSSI Values Through k-NN and Deep Learning Techniques

GPS-based positioning faces significant challenges in accuracy and reliability, especially due to environmental factors such as signal interruptions, multi-path propagation, and poor satellite visibility. This study explores using RF signal strength (RSSI) to estimate object positions, comparing different algorithms in indoor and open-air environments. For indoor localization, the Mean Absolute Error (MAE) algorithm achieved a limited 66% success rate, primarily due to RSSI fluctuations caused by signal reflections from obstacles. In open-air settings, Neural Net Fitting (NNF) outperformed Machine Learning (ML). NNF demonstrated high accuracy of approximately 94.05%, indicating effective learning and minimal overfitting. The ML model achieved 74.4% accuracy, showing less stability and overall accuracy compared to NNF. Results suggest NNF is more effective for RF-based localization, particularly in open-air environments where signal propagation is less complex.