To improve the identification and detection of radio-logical materials, we propose a hybrid supervised learning and unsupervised machine learning approach to reduce the false positive rate, increase the accuracy and throughput, and augment the capabilities of the human operators. At the end of the Phase I, we will have a machine learning algorithm that is trained to recognize a variety of nuclear materials all of which are available in-house at RMD Inc. A critical component of our approach will be the generation of our own datasets. In addition to developing and training the machine learning algorithms, RMD Inc. has a wide variety of scintillators and radiological sources in-house. To increase the available data, we will explore synthetic (software) means of distorting, or modifying the acquired data in order to increase the size of our dataset. More specifically, we propose to use a clustering algorithm for anomaly detection where there is little data (e.g., low-energy PSD), followed by a perceptron neural network for PSD, and a convolutional neural network for RIID. Our approach will improve our ability to detect radiological materials under low-signal, high noise conditions. At the end of Phase II, our algorithms will be integrated into radiation detection.