Recent advances in the machine learning field have given rise to efficient ensemble methods that accurately forecast time-series. In this paper, we use the Quantopian algorithmic stock market trading simulator to assess ensemble methods performance in daily prediction and trading. The ensemble methods used are Extremely Randomized Trees, Random Forest, and Gradient Boosting. All methods are trained using multiple technical indicators and automatic stock selection is used. Simulation results show significant returns relative to the benchmark and large values of alpha are produced from all methods. These results strengthen the role of ensemble method based machine learning in automated stock market trading.

Routing in cognitive radio networks (CRNs) necessitates a cross-layering approach. However, according to [1], CRN routing protocols proposed in literature are partially cross-layer, because the information flow is only from physical layer to network layer, e.g., about channels availabilities. In this work, we introduce a cross-layer routing protocol (CLRP), which considers both the channels that are known to be available at each node, as well as other channels that may be available. The availabilities of the latter channels are considered using a stochastic approach. CLRP computes an end to end path, and feeds the physical layer with information about which channels to sense and which nodes should perform the sensing, such that the expected route quality is enhanced. Simulation results show that CLRP outperforms other cross-layer routing protocols in terms of throughput and stability of the path being setup, and increases the probability of finding an end-to-end path.

The cognitive radio (CR) nodes in a cognitive radio network (CRN) do not have license to use specific spectrum band. Instead, they use the spectrum bands of the licensed primary users (PU) without interfering with the PU. When the PU becomes active, interfering CRs should leave to another available spectrum band within the PU's tolerable interference delay (TID). Therefore, CRN operates over wide spectrum bands which span many channels. Since each channel is typically licensed to one PU, this requires that channels be sensed separately. This adds monitoring overhead, where the CR should monitor (sense) the channel every TID, which reduces the throughput. For this reason, the node cannot monitor the whole set of channels. Deciding which set of channels to monitor affect other functions in the CRN like routing. Work done on routing in literature assumes that each node maintains a set of available channels which is obtained by sensing. Route setup decision will be made based on the available sets at all nodes. However, there may be some other available channels that the node is not aware of their availability which may enhance the routing quality metric. Also, taking into consideration only the sets of channels available at the CR nodes may preclude finding an end-to-end path. In this paper, we propose a spectrum decision framework that is complementary to the existing routing protocols. This framework is based on two objectives: 1) enhancing the route quality by sensing a few more channels at some nodes. These channels can enhance the quality by: reducing the switching time, requiring shorter sensing time, or expected to be available for longer time; 2) increasing the probability of finding a path by sensing more channels at some nodes in case the routing protocol did not find a path. Simulation results show that the proposed framework can result in enhancement that can be as high as 100% over the routing protocols that build their decisions based on the available channels at each node only.