Arc faults in DC distribution systems pose significant safety risks, including electrical fires and system failures, yet are challenging to detect due to the subtle nature of the associated signals. This study presents a method for anomaly detection using wavelet transforms and autoencoders to identify arc faults accurately. The approach utilizes Daubechies db3 wavelet to extract features from load signals, which are then used to train an autoencoder on data from normal operating conditions. The autoencoder detects faults by measuring the root-mean-square error (RMSE) between the input signal and its reconstruction. A comparative analysis was conducted between an autoencoder trained on raw signals and one trained on wavelet-filtered features. The wavelet-based autoencoder demonstrated a detection probability of 97.52% on test data with 40 arc fault regions, significantly outperforming the 57.85% detection rate achieved with raw signal data. The effect of signal normalization was examined, with z-score normalization improving detection accuracy further. In extended tests, the wavelet-filtered autoencoder achieved a 96% detection probability, demonstrating the efficacy of this approach in real-time anomaly detection for DC systems.