Stress is a common problem that affects people from all backgrounds, professions, and cultures. The physical and mental effects of stress can lead to health issues such as anxiety, depression, and heart disease. With an increasing interest in detecting stress in real time, machine learning (ML) and wearable sensors that collect biometric data offer a promising solution. This research evaluated the predictive performance of various ML models using the SWELL-KW dataset, which is designed to replicate real-world work stressors like time pressure and interruptions. The dataset features synchronized computer interaction logs, which include keyboard and mouse activity and application usage. It also collects physiological signals like heart rate and skin conductance, along with data on facial expressions and body postures. Additionally, participants provide self-reported measures of workload, emotion, and stress. We looked at several stress prediction models, including Artificial Neural Networks (ANN), Decision Trees, Support Vector Machines (SVM), and Logistic Regression. Among these, the Decision Tree classifier showed the highest accuracy at 97%, demonstrating good performance across different stress categories. However, ANN showed limited effectiveness due to class imbalance. Our classification work and comparative evaluation pointed out the strengths and weaknesses of these ML models. Lastly, the study emphasizes the potential of combining physiological data with behavioral insights using machine learning to create real-time stress detection systems that can support personalized mental health care.