Malware is malicious software designed to damage or infiltrate a computer system without the user's knowledge. Malware detection includes the process of detecting the presence of malware on the host system or determining whether the type of program is malicious or benign. Recently, machine learning (ML) algorithms have been used to detect malware effectively. Unfortunately, the core techniques require extensive feature learning, engineering and representation, which increases the computational time, error rate ratio and improves recall. The feature engineering phase of these methods can be alleviated by using more advanced ML approaches during the detection phase. In this article, we propose a gradient-boosted convolutional neural network (GB-CNN) to detect malware in Android smartphones. This proposed technique uses entropy-based feature selection technique to select relevant Android features and APKs. These selected features are fed to deep learning for classification. The classification results are then optimized by gradient boost machine learning. Comparative results show that GB-CNN outperforms other existing deep learning –based detection techniques, and is especially suitable for malware detection on Android devices, with improvements in terms of accuracy (3%), precision (1%) ,F-measure (1%), runtime (1.415SI), AUC (3.5%), recall (2%),TNR (2%),TPR (5%),FNR (15%), error rate (35%), and FPR (52%) on Android application sets. These improvements stem from optimizing the convolutional network with gradient boosting machine during the malware detection phase.