Machine Learning (ML) based anomaly detection in Smart Water Metering Networks (SWMNs) and traditional water metering networks faces significant challenges, including small dataset sizes, unlabeled data, and class imbalance. These issues often lead to high false detection rates, model overfitting, and increased model complexity. This paper aims to enhance anomaly detection in SWMNs using a ML-based Long Short-Term Memory Autoencoder (LSTM-AE) augmented with data augmentation techniques. The dataset, collected from Location A in Windhoek, Namibia, comprised of only 72 monthly water consumption data points per household, which is insufficient for effectively training ML models. To address this limitation, we employed two interpolation-based data augmentation techniques: Piecewise Cubic Hermite Interpolating Polynomial (PCHIP) and Cubic Spline, to generate synthetic data instances to expand the dataset. These techniques were selected based on the patterns and trends of water consumption data and were compared for their accuracy. The original and augmented datasets were then used to train the LSTM-AE model. The results demonstrate that, despite the limited data available in the water sector, data augmentation techniques can significantly enrich datasets, leading to substantial performance improvement with augmented data. The proposed approach greatly enhanced the accuracy of the LSTM-AE model in reconstructing data from small datasets. This study underscores the potential of data augmentation to overcome data scarcity in the water sector and enhance anomaly detection in SWMNs.