This work presents a distributed optical fiber specklegram sensor (FSS) specifically designed for the detection and localization of water leaks. The sensor analyzes specklegram images generated by a No-Core Fiber (NCF) under different water leak conditions, employing a low-cost CCD camera as an interrogation unit. To enhance the accuracy of leak detection, a convolutional neural network (CNN) model is employed to post-process the specklegram images for monitoring the different water leak conditions. The sensor demonstrates high sensitivity, accurately detecting water volumes as small as 0.1 mL. In the initial series of experiments, the sensor achieved a remarkable 100% accuracy in predicting the location of leak spots situated 1 cm apart. However, in subsequent rounds of the experiment, a slight reduction in accuracy was observed (87.5%) due to the issue of water droplet overflow across the Kapton tape used to mark the various test leak spots after multiple cycles of water addition and removal. Therefore, employing an impermeable material for the demarcation will mitigate the water droplet overflow problem. In summary, the proposed sensor offers an efficient approach for water leak detection through the application of machine learning-based specklegram analysis. The findings of this research underscore the potential of FSS as a low-cost, easily implementable, and real-time monitoring system for the detection and localization of water leaks.