In this study, we present a systematic methodology for cyberthreat detection leveraging machine learning algorithms. The process begins with data preparation, including loading, pre-processing, normalization, and splitting into training and validation sets. Feature extraction is performed using a Variational Autoencoder (VAE), reducing the dimensionality of the data to 20 features. Subsequently, feature selection techniques such as Variance Threshold Filter, KBest with Chi2 Filter and KBest with Mutual Information Filter are applied to further refine the feature space to 15 features. For model selection and evaluation, various algorithms including Logistic Regression, Linear Discriminant Analysis, Quadratic Discriminant Analysis, K-Nearest Neighbors, Extra Trees, Naive Bayes, and Linear SVC are evaluated. Through initial exploration, K-Nearest Neighbors, Extra Trees, Naive Bayes, and Linear SVC emerge as potential candidates. Hyper parameter tuning is conducted for Logistic Regression, Naive Bayes, and Linear SVC using Randomized Search CV. Further evaluation entails assessing the tuned models using multiple evaluation metrics such as accuracy, negative log loss, and ROC AUC score. ROC curves and confusion matrices are plotted to gain a comprehensive understanding of each model's performance. Based on the evaluation results, Logistic Regression and Naive Bayes are selected as the final models. Validation of the selected models is carried out on the validation set, and detailed classification reports are provided. Overall, our approach offers a structured framework for building and evaluating machine learning models for cyberthreat detection. The documentation provided throughout the process enhances transparency and facilitates comprehension of each step and decision rationale