AI based Segmented Anomaly Detection

This paper aims at targeting to solve the problem of anomaly detection by proposing a system that would be efficient in detecting anomalies and also being region-specific to narrow the target of detecting anomalies in a particular environment. The system elaborates on leveraging the power of convolutional autoencoders, LSTMs to detect anomalies of various kinds in an unsupervised manner, hence not being limited to the scope of labels of pre-trained modules. Furthermore, the architecture is enhanced by an ensemble of various such modules to improve the overall performance of the system. The obtained results are further analysed to provide region-specific based anomaly detection, this would help in prioritising and investigating those anomalies that would be of greater risk. With the region-specific anomaly detection module, anomalies in an environment are prioritised according to the risk it poses to that specific environment, hence making the system robust and adaptable to the specific environment it is working on. The final obtained results are then used to enforce the necessary contingencies, precautions and prevention methods to combat the same. The outcome of the system is to prepare humans to be take precautions and hence be vigilant to various anomalies/risks that one is subjected to. Preventing such risks would help humans become more intrepid and hopeful about their bright future.