Tamil Nadu College of Engineering

Gestational Diabetes Mellitus (GDM) is a serious health issue and could cause complications to the mother and child in case it is not identified and treated in time. Conventional diagnostic tools, including metabolomics, transcriptomics, and epigenetics, are usually not standardized and have a high degree of variation across studies. Such approaches are also hampered by non-diversity in the population of study, that restricts reliability. This research will overcome these limitations by taking real-time maternal diagnostic data and creating a powerful machine-learning-based method to determine the most impactful biomarkers to predict GDM. The flow of the work is based on the systematic preprocessing of data that guarantees quality and consistency. The Machine Learning classifiers including, Logistic Regression, Random Forest, Support Vector Classifiers (SVC), ExtraTrees Classifiers and Gradient Boosting Classifiers were used together with advanced feature selection methods Lasso Regression, SelectKBest, Recursive Feature Elimination (RFE), and SelectFromModel to identify the important biomarkers of GDM. Hyperparameter optimization and validation with successive iterations led to the optimization of the model and reliable predictions. The overall methodology of the study allowed determining the presence of critical maternal diagnostic features that influence GDM. The research offers useful information on the best biomarkers of GDM by integrating data preprocessing, feature selection, and model assessment. This strategy has potential in enhancing the early diagnosis of GDM and highlights the possibilities of machine learning in filling the gaps that are expensive to overcome in conventional diagnostic methods.

The growing use of artificial intelligence has given a breakthrough advantage to heuristic search and optimization mathematics since they have a close relationship of application so far in mathematics. Simply put, it lowers the computation costs and increases the quality of the solution by searching infinite large search space in an optimal way to aid the solution of complex mathematical optimization problems with the help of AI-based heuristic search methods. Particularly, heuristics are useful when traditional mathematical methods (e.g. brute-force search, exhaustive search) are not suitable because of combinatorial explosion. 1. Abstract: The article itself contains a survey concerning the AI guided heuristic search methods such as hereditary algorithm, simulated annihilation, molecule swarm improvement, tabu search, ant settlement advancement and artificial neural systems Each of them involves brilliant strategies that rely on the inspiration of nature, human logic or probabilistic to find the almost perfect solution to address the problem within a relatively short amount of time. Examples are genetic algorithms that rely on the principles of natural selection and evolution to find solutions in a more incremental way, and simulated annealing, which resembles physical annealing to not become trapped on a local optimum, and to be able to explore a larger space of solutions. This method is inspired by swarm intelligence where agents act according to the suggestions and ideas of their fellows, and past experiences and peer influence dynamically decide on potential solutions, so it is quite an effective approach to non-linear optimization problems. This form of heuristic is also applied in finance modeling, machine learning, networks optimization and engineering design problems, showing the ubiquity of this kind of heuristic in many mathematical domains. One of the most beneficial features of AI-based heuristic search methods is that it is capable of balancing exploration and exploitation when applied to search spaces. The algorithm can use exploration to search over a very broad range of possible outputs, and exploitation can use promising candidates and refine them to achieve high performance. A second optimization, can be made, that is based on hybrid heuristic techniques, in which various techniques are mixed.

Ai has transformed pharmaceutical sciences and altered drug discovery and delivery. It is an overview of the numerous applications of ai in drug research, which involves target discovery, and post-market surveillance. We discuss how machine learning, deep learning, and massive language models are tackling the high costs, time and failures in drug development. The paper summarises the target identification via ai, hit-to-lead optimisation, and medication repurposing, and the enhancement of personalised drug delivery. We contrast the conventional and ai-enabled solutions and find significant time (10-15 years to 5 years), and cost reductions. Such breakthrough technologies as predicting a protein structure with alphafold, generative models of de novo molecular design, and smart medicine delivery systems with ai and nanotechnology and internet of things options are mentioned. The quality of data and standardisation, interpretability of algorithms, regulation, and ethics are also discussed in the review. The last proposals are multimodal foundation models, federated learning frameworks, and safe and reliable ai deployment in the pharmaceutical industry through verified control systems.

Permanent Magnet Synchronous Motors (PMSM) are being increasingly employed in high-performance industrial environments that require precise, rapid, and efficient speed regulation. This study introduces a groundbreaking technique for controlling PMSM speed by integrating a Takagi–Sugeno Fuzzy Controller (TSFC) with an African Buffalo Optimization (ABO)-enhanced adaptive Deep Belief Network (DBN). The adaptive DBN dynamically adjusts the number of hidden neurons and fine-tunes weight parameters using ABO, while the TSFC, enhanced by the optimized DBN, addresses steady-state error through improved fuzzy rules and membership functions. By utilizing a soft-sign activation function, this approach achieves a minimal mean square error in estimating rated current. Performance evaluations indicate notable enhancements: a 25.3

Nowadays power, delay in addition to the area has to turn out to be the attribute features of any VLSI circuit. Usually, the delay of usual multipliers is high due to the number of computations, consequently; the overall speed of circuits become less, and increases power consumption. The performance of Digital Signal Processing (DSP) processors is frequently dependent on the Multiplier and Accumulator (MAC) unit, and three parameters determine it, namely power, area and speed. However, the performance of the conventional MAC is not good when the number of bits increases and also using several multiplication factors increases the power consumption. So, to reduce the compressor size for working with a higher level of bits in lower power and low area consumption, this paper proposes a new architecture for an effective MAC unit. In the proposed architecture, the Peres logic gates are applied in the third compression stage for reducing the power compression and delay. The outcomes demonstrate that the suggested design has high speed and low MAC unit area consumption. Furthermore, the increase in the compressor size is not affecting the system operations. The proposed architecture can be applied for future DSP system to get efficient performance.