Resource allocation in Kubernetes involves assigning computing resources like Central Processing Unit (CPU) and memory to containers to ensure efficient and predictable application performance across the cluster. A drawback of resource allocation in Kubernetes is that improper configuration of resource requests and limits can lead to inefficient resource utilization. In this manuscript, resource allocation in Kubernetes through a machine learning-based predictive optimization framework (ML-POF-RLK) is proposed. At first, input data is collected from the Kubernetes Dataset. The major objective is efficiently allocating Kubernetes resources for workflows to improve execution efficiency. The resource allocation scheme improves agility and predictability, enhancing workflow execution efficiency and reducing resource waste. The Interpretable Generalized Additive Neural Network (IGANN) is employed to predict the CPU, memory usage for proactive Kubernetes resource allocation. Finally Lotus Effect Optimizer (LEO) is used to optimize resource allocation. The proposed ML-POF-RLK approach is implemented in Python. The proposed ML-POF-RLK approach achieves an Montage peaks near 40 cores, epigenomics 30–35 cores; memory rises to 2000 Mi, then abruptly drops at 67 s with existing methods, like adaptive resource allocation for workflow containerization on kubernetes (ARA-WCK), effectual resource allocation in kubernetes utilizing machine learning (ERAK-ML) and DRS: a deep reinforcement learning enhanced kubernetes scheduler for micro service-based system (EKS-MBS-DRL).