This manuscript examines the implementation of machine learning (ML) algorithms for the detection of muscle activity based on Electromyography (EMG), utilizing Pynq-Z2 to achieve real-time operational efficiency. EMG signals were acquired through an Arduino platform and subsequently processed utilizing the PYNQ-Z2 FPGA board. A total of four ML algorithms, precisely k-Nearest Neighbours (KNN), Decision Trees (DT), Support Vector Machines (SVM), and Random Forest (RF), were implemented to categorize muscle activity attained from the EMG data. The efficacy of each algorithm was assessed concerning classification accuracy, with KNN achieving an accuracy of 97.50%, SVM attaining 98.00%, DT exhibiting a performance rate of 96.00%, and RF producing an accuracy of 95.00%. Feature extraction techniques were employed, specifically the Mean Absolute Value (MAV) as well as the Root Mean Square (RMS), yielding calculated average values of 2.83V and 3.13V, respectively. These findings underscore the viability of integrating Pynq-Z2 with machine learning models to facilitate efficient and precise muscle activity detection, thereby making substantial advancements in real-time, embedded system applications, including wearable health monitoring devices and prosthetic technologies.