The growing energy demands and the global shift towards Renewable Energy resources (RES), has resulted in rapid growth of the electrical power grid. With the integration of further RES into the power grid, the need for a resilient and smart power grid has never been greater. For meeting this necessity, the electrical power grid is undergoing a massive overhaul in terms of infrastructure, catalysed by the advent of Information and Communication Technology (ICT). Integration of ICT layers onto the power grid, has furthered the use of data analytic and processing tools, in the field of electrical power systems. Utilizing the plethora of data being collected from the power grids, and the modern ICT developments, such as, in the field of Machine Learning (ML), the quest to make the grid smarter and stronger continues. However, this being the infant stages of research, in ML implemented power grid protection, there is a vast variety of problems, yet to be discovered. This thesis, explores the efficacy of ML, used in tandem with Discrete Wavelet Transforms (DWT) and trained using basic statistical features of current and voltage signals in detection of a variety of disturbances in the power grid. The thesis can be summarised as follows: Identification of signal processing tool: The current and voltage signals, collected from the power system, can be utilised directly, for analysis of the characteristics of disturbances in it. However, by utilising signal processing tools, a greater insight into the signal can be obtained. DWT is one such tool, used for decomposing the signal, based on frequency, while also retaining the time domain data of the signal. The decomposition utilises special functions, called mother wavelets, and an infinite number these can be developed. A study, comparing a variety of family of these mother wavelets, yielded db11, of Daubechies family, suited for detection of disturbances in electrical signals. Disturbance signature generation: Disturbance signatures for commonly occurring events in a power system, along with rarer and harder to detect events such as HIF were simulated using PSCAD for multiple system voltages. Further, data for equipment failure related incipient faults were obtained from the PES work group of IEEE. These disturbance signals were decomposed using DWT, for training the ML module. ML implementation for detection and classification of disturbances: Using the signature of disturbances generated and obtained, supervised ML modules were trained to detect and classify the disturbances by analysing the basic statistical features, mean, standard deviation, variance, RMS, number of zero and mean crossings, Shannon entropy and energy, extracted from the DWT decomposition of the current and voltage signals. A comparison of the accuracy of, two of the common classification models, Decision Tree and Gradient Boosting, yielded Gradient boosting as more accurate at classification of the disturbance. A comparative analysis of the statistical features, also provided an insight into their contribution towards the accuracy of ML.