<p>Trust computation is crucial for ensuring the security of the Internet of Things (IoT). However, current trust-based mechanisms for IoT have limitations that impact data security. Sliding window-based trust schemes cannot ensure reliable trust computation due to their inability to select appropriate window lengths. Besides, recent trust scores are emphasized when considering the effect of time on trust. This can cause a sudden change in overall trust score based on recent behavior, potentially misinterpreting an honest service provider as malicious and vice versa. Moreover, clustering mechanisms used to filter recommendations in trust computation often lead to slower results. In this paper, we propose a robust trust model to address these limitations. The proposed approach determines the window length dynamically to guarantee accurate trust computation. It uses the harmonic mean of average trust score and time to prevent sudden fluctuations in trust scores. Additionally, an efficient personalized subspace clustering algorithm is used to exclude recommendations. We present an informal security analysis, along with experimental validation to demonstrate the resiliency of the proposed scheme against bad-mouthing, ballot-stuffing, and on–off attacks. The proposed scheme shows a competitive performance in detecting bad-mouthing attacks, while outperforming existing works with an approximately 44% improvement in accuracy for detecting on–off attacks. It maintains its effectiveness even when the percentage of on–off attackers increases and in scenarios where multiple attacks occur simultaneously. Additionally, the proposed scheme reduces the recommendation filtering time by 95%.</p>