This report covers the performance of four different machine learning models for anomaly detection using an already classified anomaly dataset from AWS. The end goal of this project is to identify the best machine learning models to use for anomaly detection. The four models: logistic regression, single-layer perceptron, multi-layer perceptron, and k-means clustering, were trained on the same dataset and their accuracy, recall, precision, and F1-scores were recorded and compared. The results of the report showed that linear classifiers such as logistic regression and single-layer perceptron performed poorly in classifying anomalies as they were not correctly learning from the data. On the other hand, models that supported non-linear data such as multi-layered perceptron performed significantly better than their linear-based counterparts. This implies that the best models to use to correctly classify anomalies are those that can handle data that is not linearly separable and as such are best suited for anomaly detection.
- Damon Lin
- Jerry Tang
- Ritvik Durgempudi
- Wesley Tam