Modern-day IT infrastructures are producing huge amounts of varied telemetry and log data, thus making real-time fault detection and predictive maintenance even more difficult. The traditional rule-based and statistical models do not perform well when it comes to high-dimensional, time-dependent, and imbalanced datasets. So, a hybrid deep learning and ensemble-based predictive framework is suggested, which works by merging Convolutional Neural Network (CNN) with Extreme Gradient Boosting (XGBoost) for the purpose of combining deep temporal feature extraction with ensemble decision optimization. The Microsoft Azure Predictive Maintenance dataset is employed for the purpose of model development and validation. Data preprocessing includes normalization, categorical encoding, and SMOTE-based balancing for the purpose of data integrity and learning stability. The experimental assessment demonstrates that the hybrid CNN+XGBoost model is more effective than the comparison models such as Random Forest, Decision Tree, and Logistic Regression, with an accuracy (ACC) of 98.75%, precision (PRE) of 0.987, recall (REC) of 0.993, F1-score (F1) of 0.990, and an AUC of 0.994. These results highlight the strength, scalability, and excellent predictability of the model. Through the combination of deep learning and an ensemble approach, the framework becomes a robust, adaptive, and smart model when it comes to applying proactive maintenance and operational optimization within the IT industry.

Growth in stored data and request volume turns latency degradation into a recurring condition of long-lived server systems. A frequent response is to add hardware capacity, which raises operating expenditure while the underlying defect persists. This article treats server-side profiling as a proactive engineering practice and organizes it into a four-stage procedure: localization of the layer that holds the delay, selection of a diagnostic instrument according to whether the source is evident, prioritization of findings by their share of total execution time and by call frequency, and targeted remediation followed by re-measurement. The evidence comes from two production cases on a PHP and PostgreSQL platform, documented through a descriptive multiple-case design. The first case concerns an inventory listing endpoint whose response time had grown to roughly 10 seconds. Query plan analysis revealed a recursive traversal within the main statement, incomplete index coverage for join and filter predicates, a hash operation spilling to temporary files, and expensive computed columns evaluated on rows that pagination would discard. Response time fell to roughly 500 milliseconds after four targeted changes. The second case concerns bulk warehouse intake, where a call-level profiler attributed the delay to many small recalculations repeated after every saved item, and deferring those operations into a single pass at the end of the request removed the redundancy. The cases occupy opposite ends of a diagnostic spectrum covered by one procedure, and the article argues that added capacity conceals both classes of defect while leaving them in place. The evidence base is limited to two cases from a single platform, so the findings support the transferability of the procedure rather than a statistical claim about effect size. The practical implication for engineering teams is to capture baselines for critical endpoints while they still perform acceptably, so that later degradation becomes a comparison against a known quantity rather than an incident.

Additive manufacturing (AM) has transformed modern manufacturing by enabling the layer-by-layer fabrication of complex components directly from digital models. However, variations in process parameters, material behavior, thermal conditions, and machine dynamics can significantly affect product quality, dimensional accuracy, and process stability. This review examines advanced closed-loop control techniques for additive manufacturing, emphasizing real-time monitoring, feedback control, and intelligent process regulation. The AM process, major printing technologies, and conventional closed-loop control concepts are first discussed to establish the fundamental framework. In-situ monitoring techniques, including imaging and acoustic sensing, are examined for real-time defect detection and process validation. Advanced control strategies such as Proportional-Integral-Derivative (PID) control, Adaptive Control, Gain Scheduling, Model Reference Adaptive Systems (MRAS), and fuzzy logic-based control are then reviewed. These approaches enable dynamic adjustment of printing parameters in response to process deviations, thereby improving manufacturing consistency and reducing defects. The review also highlights an integration of AI with closed-loop systems to support predictive and autonomous control. Finally, current challenges and research opportunities are identified, emphasizing intelligent sensing, adaptive algorithms, and data-driven control for reliable and autonomous additive manufacturing.

Stock market prediction is a complex computational problem because financial prices are influenced by nonlinear patterns, historical dependencies, market trends, and rapidly changing information. Conventional statistical approaches often face difficulty in representing these interacting characteristics, motivating the adoption of machine learning, deep learning, sentiment analysis, and reinforcement learning techniques. This research and review paper examines intelligent approaches for stock market prediction and decision-making through a structured synthesis of the provided literature. The study comparatively analyzes regression-based models, support vector machines, machine learning algorithms, recurrent architectures, sentiment-enhanced prediction, and reinforcement learning approaches. Particular attention is given to the methodological implications of feature selection, temporal dependencies, model comparison, and the integration of heterogeneous information. The reviewed studies indicate that no single predictive model is universally optimal; rather, model effectiveness depends on the characteristics of the dataset, prediction horizon, selected features, and evaluation methodology. The analysis further proposes an intelligent prediction and decision-making framework integrating market data, sentiment information, advanced predictive models, confidence estimation, and decision rules. The resulting framework provides a conceptual basis for transforming price prediction into an actionable decision-support process while recognizing the limitations associated with market uncertainty, data quality, model instability, and generalization.