A tremor is an involuntary shaking or trembling movement that can be caused by various factors, including neurological diseases, stress, and certain medications. Fatigue, a normal byproduct of exertion over an extended period, may also be a contributing factor. This study introduces a novel framework for the classification and quantification of tremor-related tasks, incorporating effortful, postural, and resting activities. The tremor data collected from participants who wore an accelerometer on a finger were employed to create a model for assessing and forecasting occurrences. During the experimental session, the volunteers performed various movements. Two scenarios were used to calculate the duration of tasks: in the first scenario, tasks were categorized into resting, effort, and postural activities; in the second, postural tasks were excluded due to their limited occurrence. We applied multiple machine learning algorithms, including k-nearest neighbors, decision trees, bagged ensembles, and support vector machines, using statistical features extracted across various window lengths (128, 256, 320, and 512 samples). We calculated the duration of effortful activity using a windowing approach and compared classification performance across finger, wrist, and combined wrist and finger data. Support vector machines achieved the highest accuracy: 90.9% for the two-class scenario and 90.7% for the three-class scenario with a 320-sample window. This study presents a systematic methodology for tremor detection, classification, and quantification, emphasizing the optimal placement of sensors and machine learning approaches. The results offer a scalable framework for clinical monitoring, objective assessment of tremor severity, and personalized intervention planning.