Systematic Literature Review of Time Series Clustering and Deep Learning for Dynamic Data Prediction

The increasing complexity of time-dependent data has created a growing need for analytical approaches capable of identifying temporal patterns and generating reliable predictions. Time series clustering provides a means of discovering groups of similar temporal patterns, while deep learning enables the representation of nonlinear relationships and temporal dependencies. However, the integration of these approaches with dynamic modeling remains methodologically diverse and requires systematic synthesis. This study conducted a Systematic Literature Review (SLR) to identify, classify, and synthesize research on time series clustering and deep learning for dynamic data prediction and pattern recognition during 2016–2025. The review followed the PRISMA framework and searched Google Scholar, Scopus, DOAJ, and Dimensions AI using predefined search terms and eligibility criteria. From 10,950 identified records, 2,450 duplicates were removed, 8,500 records were screened, and 400 full-text articles were assessed for eligibility, resulting in 80 studies included in the qualitative synthesis. The findings indicate an evolution from separate applications of clustering and deep learning toward integrated approaches combining clustering, temporal representation learning, and predictive architectures such as RNN, LSTM, CNN, and Transformer-based models. Clustering is primarily used for pattern discovery, segmentation, and preprocessing, whereas deep learning is applied to model nonlinear temporal dependencies and prediction tasks. The review also identifies challenges concerning computational complexity, data requirements, interpretability, scalability, hyperparameter configuration, and reproducibility. These findings provide a structured synthesis of methodological developments, relationships, limitations, and research gaps in integrated approaches for dynamic data prediction and pattern recognition.