The importance of cloud technologies will only grow, and ensuring scalability is dependent on the security of such systems. This work analyzes log files and proposes a framework to detect anomalies in such files. Light gradient boosting machine (LightGBM) is utilized for predicting anomalies, while a natural language processing method called Term Frequency-Inverse Document Frequency is used for data preprocessing. The main contribution of the work is the proposition for a novel version of the firefly algorithm towards hyperparameter optimization of LightGBM for the specific problem of anomaly detection in log files. The proposed method is validated against other state-of-the-art metaheuristic optimizers. Experimental outcomes demonstrated exceptional performance of the best produced models, attaining accuracy rates of nearly 98.53% in this scenario.