Toward a prediction approach based on deep learning in Big Data analytics

• Published in: Neural computing & applications 2023-03, Vol.35 (8), p.6043-6063
• Main Authors: Haddad, Omar, Fkih, Fethi, Omri, Mohamed Nazih
• Format: Article
• Language: English
• Subjects: Artificial Intelligence; Big Data; Cloud computing; Computational Biology/Bioinformatics; Computational Science and Engineering; Computer Science; Data analysis; Data Mining and Knowledge Discovery; Datasets; Deep learning; Empirical analysis; Image Processing and Computer Vision; Original Article; Probability and Statistics in Computer Science; Unstructured data
• ISSN: 0941-0643; 1433-3058
• DOI: 10.1007/s00521-022-07986-9

Nowadays, cloud computing plays an important role in the process of storing both structured and unstructured data. This contributed to a very large data growth on web servers, which has come to be called Big Data. Cloud computing technology is adopted in many applications, perhaps the most important of which are social networking applications, e-mail messages, and others, which represent an important source of data through the process of communication between web users. Thus, these data represent views and opinions on various topics, which can help businesses and other decision makers in making decisions based on future predictions. To achieve this goal, several methods have been proposed. Recently, it relies on the use of deep learning as a tool for processing large volumes of data due to its high performance in extracting predictions from the opinions of web users. This paper presents a new prediction approach based on Big Data analysis and deep learning for large-scale data, called PABIDDL. The infrastructure of the proposed approach is focused on three important stages, starting with the reduction of Big Data based on MapReduce using the Hadoop framework. In the second stage, we performed the initialization of these data using the GloVe technique. Finally, the text data were classified into advantages and disadvantages poles depending on CNN deep learning approach. Also, we conducted an empirical study of our proposed approach PABIDDL and related works models on two standard datasets IMDB and MR datasets. The results obtained showed that the best performance is given by our approach. We recorded 0.93%, 0.90%, and 0.92% as an accuracy, a recall, and an F1-score, respectively. Furthermore, our approach reached the fastest response time.