Fog computing is a geo-distributed computing network; trust must be established between fog nodes for secured data communication. Decision-making can be aided by each nodes' ability to foresee the behaviours of other nodes according to trust value calculated, either directly or indirectly. Especially in indirect trust value calculation, decision to accept the recommending nodes value is critic due to malicious fog nodes which cause internal attacks like self-promotion attack, bad-mouthing attack, and on-off attack. This paper proposes a reputation based trust management system (RTMS) for malicious fog node detection among the recommending fog nodes using geometric probability distribution on multi-dimensional attributes. This RTMS keeps track of the past records of the fog node to calculate the reputation value of each recommender using weighted geometric mean in the fog environment. The proposed technique successfully detects the malicious fog nodes, that are possible among the recommending fog nodes, and allows only the trustworthy node's recommendation for indirect trust calculation, thereby eliminating malicious nodes. The simulation shows that RTMS outperforms the previous work in terms of suitability and security.