The advent of information and communication technology has revolutionized the application of smart home through which the residents remotely establish superior control over the smart devices. The users and the smart devices interact over an insecure communication channel that may help an illegitimate user to gain access on the data disseminated by the smart devices, resulting in high exploiting attacks like privileged insider attack, smart device and gateway node impersonation attack, and smart node capture attack in the smart home. Thus, a secure remote authentication scheme that permits only legitimate users to gain access to the smart devices based on supervised learning becomes essential. In this article, Dynamic Forest of Random Subsets-based supervised hash signature (DFORS-SHS) scheme is proposed for robust user authentication in smart homes. This proposed DFORS-SHS scheme concurrently performs the signed generation of random subsets such that each element of a signature is used for the generation of successive random subset elements for authentication. It utilized the strength of revealed secret keys through chaining approach that bundles the random subset generation and signature. This supervised learning approach only permits the signer of the original message to potentially generate a random subset and prevents the malicious intruder from gaining the access of the smart devices. The formal verification and the security investigation of the proposed DFORS-SHS scheme are conducted using the Burrows-Abadi-Needham (BAN) and AVIPSA toolkit for determining communication cost, communication time, computation time, and energy consumptions incurred during the authentication process. It can be stated that, the communication cost and storage costs of the proposed DFORS-SHS scheme was also identified to be comparatively minimized than the baseline smart home authentication schemes. © 2021 John Wiley & Sons Ltd.