The rapid evolution of smartphone technology has led to a surge in cyber-attacks specifically targeting smart devices. Of particular concern is the prevalent practice among users to grant permissions to arbitrary programs without due consideration, thereby undermining the efficacy of the authorization system. While numerous malware detection methods have been proposed, they often exhibit limitations, including inadequate identification and detection rates. To address these pressing challenges, this paper presents a novel approach called Convolutional Neural Network-Based Adaptive Red Fox Optimization (CNN-ARFO) aimed at discerning between normal and malicious malware applications on Android smartphones. The methodology involves the pre-processing of the dataset, wherein the Minmax approach is employed to effectively normalize features. Subsequently, during extraction, the CNN-ARFO method meticulously scrutinizes malicious APKs, curating a representative selection of normal apps to extract essential characteristics crucial for identifying malware operations. The foundation of the proposed method lies in harnessing the Adaptive Red Fox Optimization (ARFO) technique in conjunction with Convolutional Neural Networks (CNNs) to achieve precise and reliable malware detection. The CNN-ARFO approach is comprehensively evaluated to assess its efficacy, and its performance is compared against other state-of-the-art malware detection techniques, including DANN, DBN, MLBS, and DE-CNN. A diverse set of evaluation metrics, including accuracy, precision, recall, and f-measure, are employed for a thorough analysis. Experimental results indicate the superiority of the CNN-ARFO method, demonstrating outstanding performance with accuracy, precision, recall, and f-measure values of 99.20%, 97.45%, 91.23%, and 98.43%, respectively.