This paper presents an Imagined digit classification system based on Electroencephalogram (EEG) signals, utilizing machine learning models. The purpose of Imagined speech is to provide the communication medium to the people who are unable to produce speech due to physical or neurological impairments. In such cases, speech can be obtained by measuring the electrical signals recorded from brain by placing the electrodes on the scalp. The recorded brain patterns can be analyzed to interpret the Imagined Speech. This research uses a publicly available dataset that consists of 2-s segments of brain signals, captured with the visual stimulus of a digit and non-digit. The proposed approach involves signal pre-processing techniques to eliminate noise from raw EEG signal, followed by feature extraction in time, frequency and time-frequency domain. Various supervised machine learning models are then employed to classify the imagined digit into two classes, digit and non-digit. The experimental results show that Random Forest classifier is able to perform classification with a maximum accuracy of 90.57%. Among the different feature domains, features extracted from time-frequency domain yielded the highest accuracy.