In this paper, we propose two models for scaling the transaction throughput in Proof-of-Work (PoW) based blockchain networks. In the first approach, a mathematical model has derived for optimal transaction throughput for PoW based longest chain rule blockchain. In this approach, the blockchain Peer-to-Peer (P2P) network is considered as Erdös-Rényi random network topology. This approach is however limited by the block creation rate, the results suggest that the rate beyond an optimal point can result in unfairness in the system. The second approach is a new consensus protocol proposed by considering the ledger as a Directed Acyclic Graph (DAG) called blockDAG instead of a chain of blocks. In this framework, we follow a two-step strategy that makes the system robust enough to handle the double-spend attacks. The first step involves the development of an unsupervised learning graph clustering algorithm for separating the blocks created by an attacker. In the second step, the attackers blocks are eliminated and the remaining blocks are arranged in topological order by honest clients which makes the blockDAG system suitable for smart contract applications found in Internet of Things (IoT) services. The Simulation results demonstrate a significant improvement in the transaction throughput compared to bitcoin.