In digital communications, block ciphers are vital for ensuring data confidentiality and integrity. Identifying these ciphers from intercepted ciphertexts is vital for cybersecurity; however, traditional methods often assume prior knowledge of the encryption algorithm. This study explores a hybrid K-nearest neighbor random forest (KNNRF) framework for block-cipher recognition, addressing these limitations using ensemble learning. Key innovations include an adaptive feature selection method that enhances efficiency and accuracy and a dynamic ensemble approach combining K-Nearest Neighbors (K-NN) for localized recognition with Random Forest (RF) for global analysis. Using only two statistical features—Approximate Entropy (ApEn) and Discrete Fourier Transform (DFT) P-values—the model achieves up to 99% accuracy for smaller datasets (1 KB and 8 KB) and maintains 98% accuracy for larger datasets (64 KB and 256 KB). This represents a significant improvement over traditional methods, such as the multilayer perceptron (MLP), which achieved only 46% accuracy. The efficiency and minimal feature requirements of the KNNRF framework make it ideal for real-time applications in cybersecurity, demonstrating that optimized ensemble learning can effectively bridge the gap between theoretical advancements and practical applications.