Emotion recognition is a cornerstone of affective computing, offering crucial insights into human states for a myriad of applications. Physiological signals, such as electrodermal activity (EDA), provide objective measures of emotional arousal and valence. However, transitioning emotion recognition systems from controlled laboratory settings to unconstrained real-world environments presents significant challenges, primarily due to increased signal noise and the complexity of preprocessing raw physiological data. This study addresses these limitations by introducing a novel machine learning approach for emotion recog-nition that directly utilizes raw EDA signals, thereby eliminating the reliance on complex and computationally intensive preprocessing steps typically found in existing methods. Through comprehensive experimentation, our proposed method demonstrates competitive classification accuracies for both emotional dimensions from raw EDA. This research holds substantial significance as it pioneers real-world emotion recognition directly from raw EDA signals, marking a pivotal advancement in the field. By eliminating the necessity for intricate preprocessing pipelines, our method offers a highly practical and simplified solution, paving the way for more accessible and practical affective computing systems in real-world environment.