The extensive applications of wires and cables in industries are a result of the rapid advancements in electronic device technology. Despite the established use of wireless technology, safety concerns have sustained the reliance on cables. Detecting soft faults is a challenging task due to the fact that they create reflections that are not easily distinguishable. Time and frequency domain reflectometry are commonly used for fault detection. Here, we propose a stable and effective method for estimating transmission line (TL) impedance using frequency domain reflectometry with neural network models. This method not only identifies the location of soft faults but also provides an impedance profile across the TLs. The performance of the proposed method is verified using simulated and experimental data. Moreover, we address the limitations of the Born approximations, which become ineffective for lossy TLs. The proposed method takes into account the loss information of the TLs, enabling accurate estimations of both the real and imaginary parts of the complex impedance values.