This study presents a lightweight unsupervised anomaly detection approach for cloud network traffic using Principal Component Analysis and Isolation Forest. The objective is to examine the effect of dimensionality reduction on anomaly detection performance in high-dimensional network traffic. The proposed method applies PCA to reduce 38 numerical features to 20 principal components, retaining 91.97 percent variance after standardization. Isolation Forest is then trained using a contamination value of 0.25 derived from the dataset distribution. A Random Forest with SMOTE baseline is used for comparison. The model is evaluated on the NSL KDD benchmark dataset and achieves 72.87 percent accuracy and 86.75 percent attack precision with an average inference time of 12.4 milliseconds per instance. Comparative analysis demonstrates that the proposed framework outperforms a supervised Random Forest with SMOTE baseline in terms of F1-score while maintaining real-time performance. The results demonstrate that the suggested methodology offers an effective trade-off between the performance of anomaly detection and computational efficiency, and can be possible in the context of lightweight cloud traffic monitoring in controlled settings...