Stock market forecasting involves predicting fluctuations and trends in the value of financial assets, utilizing statistical and machine learning models to analyze historical market data for insights into future behavior. This practice aids investors, traders, financial institutions, and governments in making informed decisions, managing risks, and assessing economic conditions. Forecasting financial markets is difficult due to the intricate interplay of global economics, politics, and investor sentiment, making it inherently unpredictable. This study introduces a Deep Learning based Expert Framework for Stock Market forecasting (Portfolio prediction) called DLEF-SM. The methodology begins with an improved jellyfish-induced filtering (IJF-F) technique for preprocessing, effectively analyzing raw data and eliminating artifacts. To address imbalanced data and enhance data quality, pre-trained convolutional neural network (CNN) architectures, VGGFace2 and ResNet-50, are used for feature extraction. Additionally, an improved black widow optimization (IBWO) algorithm is designed for feature selection, reducing data dimensionality and preventing under-fitting. For precise stock market predictions, integrate deep reinforcement learning with artificial neural network (DRL-ANN) is proposed. Simulation outcomes reveal that the proposed framework achieves maximum forecasting accuracy, reaching 99.562%, 98.235%, and 98.825% for S&P500-S, S&P500-L, and DAX markets, respectively.

Purpose: The purpose of this study is to analyze the effects of nonlinear thermal radiation, Darcy–Forchheimer dissipation and Joule heating on the flow of a power-law conducting fluid over a stretching porous wedge embedded in a Darcy–Forchheimer porous medium. The study also aims to develop and validate a deep learning framework based on the Levenberg–Marquardt backpropagation algorithm to accurately predict velocity and temperature distributions in complex nonlinear transport phenomena. Design/methodology/approach: The governing nonlinear partial differential equations are transformed into nonlinear ordinary differential equations using a second-order local non-similarity approach. The resulting boundary value problem is solved numerically using MATLAB’s bvp4c solver. A deep learning model using the Levenberg–Marquardt algorithm, gradient descent optimization and a hyperbolic tangent sigmoid activation function is used to predict velocity and temperature profiles. Findings: The deep learning predictions show excellent agreement with numerical solutions, with absolute errors of order 10−4–10−5 and very low mean squared errors. Increasing radiation and temperature ratio parameters significantly modifies the thermal boundary layer, causing non-monotonic temperature behavior near the wedge surface. Darcy dissipation reduces fluid temperature, while Forchheimer dissipation increases it. Originality/value: This study combines nonlinear thermal radiation and Darcy–Forchheimer effects within a second-order local non-similarity framework, alongside a Levenberg–Marquardt-based deep learning model. The hybrid approach delivers highly accurate predictions for complex porous media flows, demonstrating the power of deep learning beyond traditional numerical methods.

Purpose: The study aims to evaluate and predict the influence of different wall motions on heat transfer rate and fluid friction in unsteady magnetohydrodynamic flow of a nonNewtonian Cross fluid. In addition, it seeks to develop a hybrid numerical–artificial intelligence framework to accurately predict the flow field, thermal distribution and concentration field, enhancing predictive accuracy and computational efficiency in thermal engineering applications. Design/methodology/approach: The governing nonlinear equations for unsteady MHD flow of a nonNewtonian Cross fluid are solved using the finite difference method, and the resulting data set is used to train an artificial neural network to predict the flow field, thermal distribution and concentration field. Findings: A detailed case study of wall kinematics demonstrates that steady wall motion provides a baseline thermal response, whereas pure oscillatory motion enhances heat transfer by 29.4% due to intensified near-wall mixing. Ramp-oscillatory motion yields the maximum thermal enhancement, increasing heat transfer by 73.2%, while phase-shifted oscillations improve thermal transport by approximately 65% and concurrently reduce fluid friction, indicating superior thermo-hydrodynamic performance. Mass transfer remains insensitive to wall kinematics. The ANN predictions achieve high accuracy with absolute errors in the range of 10−4 to 10−5. Originality/value: The study presents a hybrid numerical–artificial intelligence framework for predicting heat transfer and fluid friction in unsteady MHD flow of a nonNewtonian Cross fluid. This approach provides an efficient and accurate strategy for analyzing the effects of complex wall motions, offering insights for thermal optimization and predictive control in oscillatory nonNewtonian flow systems.

Serving large language models (LLMs) on cloud spot and preemptible instances reduces costs by 60 to 90 percent compared to on-demand pricing, but unpredictable instance preemptions cause request failures, KV-cache state loss, and degraded user experience. We present MorphCloud-LLM, an elastic LLM serving system designed to achieve the reliability properties of on-demand serving at spot-instance pricing. MorphCloud-LLM integrates three synergistic components: (1) an asynchronous incremental KV-cache checkpointing engine that streams only delta state to disaggregated persistent storage with less than 3% throughput overhead, enabling sub-second KV-cache delta streaming and reconstruction for KV-cache sizes up to 32 GB on replacement instances (total end-to-end migration latency: 1390 ms); (2) a gradient-boosted preemption prediction model trained on spot market telemetry that achieves 89% recall at a 30-s prediction horizon, providing sufficient lead time for proactive migration before forced eviction; and (3) a speculative decoding continuity engine that offloads draft model token generation to on-demand fallback nodes during migration windows, bounding the user-visible interruption to a sub-second buffering pause. MorphCloud-LLM is deployed and evaluated on AWS and GCP using LLaMA-70B and Mixtral-8x7B across 521 trace-injected preemption events, achieving up to 76% cost reduction under active-serving accounting (69.8% for LLaMA-70B; 67% including warm standby fallback capacity) with only 2.1% p99 latency overhead and zero dropped requests. Extensive ablation studies confirm the contribution of each component to overall system resilience. Note that preemption events are reproduced via a trace-driven simulation framework built on empirical AWS and GCP spot interruption traces rather than fully uncontrolled live production preemptions. Production generalizability under uncontrolled preemption—including simultaneous multi-node failures, network congestion, storage contention, and replacement-instance scarcity remains subject to future validation in sustained live deployments.