Effective epilepsy management requires accurate and timely seizure detection. Traditional models are often patient-specific, limiting their generalizability across different individuals and effort for physicians is increased by enormous labeling. This work proposes a patient-non-specific approach for seizure detection using multichannel EEG signals. The method involves multiband feature fusion, where EEG signals are decomposed into distinct frequency bands using wavelet transform to capture diverse characteristics of seizure activity. Statistical and entropy-based features are extracted from these bands, providing a detailed representation of temporal and frequency-domain dynamics. These features are then combined into a unified set to enhance the robustness of the detection model. Random Forest and AdaBoost classifiers are applied to distinguish between seizure and non-seizure states. The proposed approach has achieved an accuracy of 94.73%, precision of 96.20%, sensitivity of 78.00%, and specificity of 99.27%. This model demonstrates strong potential for clinical use, offering a scalable and generalized solution for seizure detection that can be applied across a wide range of patients.