The research aims to optimize large-scale data analytics in cloud computing by incorporating DL techniques for proficient workload distribution. The implemented model minimizes execution time and resource utilization, improving scalability and performance in high-volume cloud applications.

The exponential growth of big data, cloud computing, and IoT demands competent, scalable, and protected processing frameworks. Traditional clustering and workload distribution methods resist optimization and real-time compliance. Security vulnerabilities in cloud-based data environments delay widespread implementation. Cross-cloud reproducibility remains a challenge, resisting seamless interoperability. The research improves scalability, effectiveness, and security through AI-driven optimization and security-by-design frameworks for real-time data analytics. The implemented approach overcomes the limitations of the research and provides better performance than traditional methods.

The cloud workload dataset for scheduling analysis was collected from the open-source Kaggle website: https://www.kaggle.com/datasets/zoya77/cloud-workload-dataset-for-scheduling-analysis. The dataset is intended for use in DL-based computational techniques for workload scheduling optimization and large-scale cloud workload analytics. By recording job execution data, resource use, and scheduling tactics, it replicates real-world cloud settings. In cloud computing frameworks, the dataset might be used to assess scheduling efficiency, optimize task allocation, and reduce execution delays.

The procedure of cleaning, transforming, and organizing raw data to improve its value and usability is data preprocessing. In cloud computing, intelligent preprocessing techniques handle missing values and apply Z-score standardization to ensure reliability. The steps improve workload distribution and avoid resource allocation. Feature selection and dimensionality reduction promote optimized computational effectiveness. By incorporating the techniques, the implemented model guarantees increased system performance and resource utilization.

The extensively utilized data standardization technique was Z-score standardization [14]. In the context of the research, Z-score standardization is applied to preprocess large-scale datasets collected from cloud platforms, IoT devices, and enterprise databases. Standardizing features before feeding into the proposed model ensures consistent data scaling, preventing dominant features from skewing workload distribution. This contributes to optimizing computational resources, reducing CPU and memory consumption, and enhancing overall system throughput in high-volume cloud computing environments. Equation (1) presents the z-score standardization for a feature  x′.

(1)

The original value is denoted by x, the normalized value by x′, the mean value of the feature x by mean(x), and the standard deviation of x by std(x). Compared to other standardization techniques, Z-score standardization has the notable benefit of being resilient against outliers. Extreme values that deviate greatly from the majority of data points are known as outliers, and the values have the potential to skew statistical characteristics and impact how workload is distributed in cloud computing. Applying Z-score normalization ensures balanced scaling, preventing such distortions and enhancing the efficiency of the proposed technique.

Ensuring data completeness is vital for accuracy and reliability in large-scale data analytics. Missing data, caused by factors like manual errors, sensor failures, or inconsistent collection from cloud and IoT sources, can reduce computational efficiency. The proposed model addresses the limits through robust preprocessing, categorizing missing data into three types to enhance DL-driven scheduling in a cloud environment.

By incorporating strong preprocessing techniques in the implemented technique, the framework ensures high-quality data input, thus enhancing the evaluated metrics. The capability to handle missing values efficiently improves model performance, supporting scalable and adaptive workload scheduling in cloud computing environments.

The proposed ASB-Dynamic-CapsNet integrates the ASB with a Dynamic CapsNet to improve workload distribution in cloud computing. The ASB efficiently searches for optimal scheduling solutions, reducing resource consumption, while the Dynamic CapsNet conserves spatial relationships in compound data patterns for precise task execution. The hybrid approach guarantees adaptive learning, dynamically adjusting to workload deviations. By leveraging the included strengths of DL, the framework achieves better scalability, efficiency, and task prioritization in cloud-based data analytics.

CapsNet was intended to encode spatial hierarchies by utilizing vector-based neurons called capsules. The concept started with Transforming Auto Encoders (TAEs), which need external transformation matrices. Later, Dynamic Routing CapsNet was introduced for internal pose estimation, followed by matrix capsules that improved accuracy while reducing computational complexity. CapsNet [15] optimization focuses on enhancing efficiency and scalability. Dynamic routing mechanisms develop how capsules interrelate, while metaheuristic algorithms enhance routing decisions and workload circulation. The optimizations help in running large-scale data analytics while reducing execution time and resource consumption. Figure 2 shows the general architecture of CapsNet.

CapsNet with dynamic routing provides a viable option to perform large-scale data analytics while preserving spatial relations in data. Unlike traditional approaches, CapsNet uses vector-based representations to encode information hierarchically for better feature extraction and classification accuracy. The activation function applied to capsules is a non-linear squashing function, ensuring that the output vector maintains an interpretable magnitude normalized in the range of [0, 1], as represented in equation (2).

(2)

Where T_i represents the total input to capsule i; it guarantees that greater vector magnitudes relate to an increased probability that specific features in the data were present, which enhances the prediction accuracy made by computation models in the cloud. The dynamic routing mechanism connects the lower-layer capsules with those at higher layers and adjusts weights utilizing an iteratively derived consensus process, as stated in equation (3).

(3)

Where d_ji presents coupling coefficients, and X_ji is a transformation matrix that ensures optimal feature alignment. The adaptability of any cloud-based analytics system is hence greatly enhanced by this mechanism, especially in those considered real-time data-processing and hierarchical feature-learning setups. Dynamic routing mechanisms based on CapsNet provide an influx of feature retention, minimal redundancy in computation, and realistic prediction accuracy development to the large-scale data analytics workflows running on cloud computing frameworks. It lends itself to computationally intensive optimization schemes in a cloud paradigm to enable feasible data-driven decision-making across distributed systems.

The ASB optimizer is an evolutionary computation technique inspired by the nesting activities of satin bowerbirds. In the context of cloud-based large-scale data analytics, ASB is engaged in task scheduling and resource optimization. The effectiveness of a cloud computing system relies on effective workload distribution. The attractiveness of each solution is assessed using a fitness function, which quantifies its efficacy in optimizing cloud performance. The probability of selecting a solution was represented in equation (4).

(4)

Where fiti represents the efficiency scores of the j th scheduling strategy, and the total number of persons was denoted by NP. The fitness function is computed using equation (5).

fiti={(1/(1+f(w_j)) if f(w_j) ≥0, 1+|f(w_j)| if f(w_j) < 0) (5)

Where f(w_j) represents the execution time and resource utilization associated with the

jth scheduling strategy. Integrating ASB with CapsNet enhances the load-balancing and task-execution efficiency of the framework. ASB employs dynamic resource allocation for scheduling, while CapsNet maintains spatial data relationships for better feature learning. The fusion of metaheuristic optimization and DL yields a scalable, efficient cloud computing model, which outperforms traditional scheduling methods and provides superior advantages to large-scale data analytics.

Python 3.8 on a Windows 11 system was used to implement the proposed method, which is operational with an Intel Core i7 processor. DL frameworks, such as TensorFlow and PyTorch, were used for training and optimization of the ASB-Dynamic-CapsNet model. Data preprocessing, which includes handling missing values, feature extraction, and transformation, was accomplished using NumPy and Pandas. Performance evaluation and visualization of the experimental results were done with Matplotlib and Seaborn.

The correlation matrix is used to quantify the relationships between the metrics and provide insight into the mutual influence. The research eventually enables better decision-making for dynamic scheduling and resource allocation strategies to maximize system performance and efficiency. Figure 3 represents the correlation matrix, where the strength and direction of relationships between different metrics such as CPU_Utilization (%), Memory Consumption (MB), Schedular_Type, and Job_Priority are visually represented using a heat map.

Efficient CPU utilization is an essential factor in optimizing workload execution and resource management in cloud environments. Monitoring CPU usage patterns over time provides insights into task execution efficiency and system performance. Figure 4 presents a time-series analysis of CPU utilization, illustrating fluctuations in processing demand across various tasks.

The performance of ASB-Dynamic-CapsNet was evaluated using key evaluation metrics. Training accuracy measures learning efficiency across epochs. CPU utilization evaluates workload distribution, ensuring optimal processing. RAM utilization analyzes memory efficiency for large-scale task management. These metrics collectively exhibit the model’s scalability and resource optimization in cloud computing.

Optimizing large-scale data analytics in cloud computing requires efficient workload distribution and adaptive scheduling strategies. The implemented ASB-Dynamic-CapsNet model improves scheduling performance by combining DL with metaheuristic optimization, outperforming existing models [16], such as RL, RNN-LSTM, DQN, and DRL-LSTM, in adaptability and resource efficiency. Unlike these approaches, which can struggle with dynamic workload variations and higher computational overhead, ASB-Dynamic-CapsNet ensures optimal resource utilization while maintaining task execution efficiency. The findings highlight its potential for improving cloud-based analytics and minimizing system bottlenecks. However, challenges such as computational complexity and scalability remain. Future research should focus on reducing processing overhead, integrating federated learning for decentralized cloud environments, and strengthening security to enhance real-time applicability in large-scale cloud computing.