The cyber physical system (CPS) involves incorporation of physical system into real-time software management in cyber world where network interrelates the data exchanges. The existing anomaly techniques have local optima issue and less convergence rate. This research proposes an inertia weight-based elephant herding optimization and support vector machine (IWEHO-SVM) for anomaly detection. The one-hot encoding (OHE) and standardization are applied for preprocessing the CIC-IDS2018, CIC-IDS2017, and KDD99 datasets. By using IWEHO, best features are selected and classified by SVM. To balance exploration and exploitation of EHO, inertia weight is involved to EHO that prevents model from local optima issue. In early iteration, inertia weights are higher which enables population for faster rate convergence. The SVM is used with various kernel function which enables to handle both nonlinear and linear data. The SVM is efficient in separation margin scenario that are less prone for overfitting particularly in higher-dimensional space. The IWEHO-SVM performance is calculated with the metrics of precision, f1-score, accuracy, and sensitivity. The IWEHO-SVM obtained 99.62%, 99.36%, and 99.55% accuracy for CIC-IDS2018, CIC-IDS2017, and KDD99 datasets, respectively, when compared to convolution neural network-bidirectional long short-term memory (CNN-BiLSTM).