Search

Every year, maritime accidents cause severe damages not only to humans but also to maritime instruments like vessels. The authors of this work therefore propose a machine learning-based maritime accident prediction system that can be used to prevent maritime accidents from happening by predicting and interpreting the accidents. This work overcomes the limitations of the existing works that lack practicability in the sense that the ex-post analyses are conducted to suggest accident prevention strategies but maritime accidents are not analyzed holistically. Using extensive literature reviews and expert interviews, a large number of risk factors associated with maritime accidents are identified, and related data are collected and utilized in the work. Throughout variable selection, data retrieval, hot-spot identification, and the maritime accident prediction model construction process, various machine learning algorithms are exploited in order to construct an organized system. In addition, interpretations for the resulting accident predictions are given using interpretable machine learning algorithms so as to provide explainable results to users. Finally, the proposed system is evaluated using a SERVQUAL model and proves its effectiveness in real-world applications.

Cryptocurrency has recently attracted substantial interest from investors due to its underlying philosophy of decentralization and transparency. Considering cryptocurrency’s volatility and unique characteristics, accurate price prediction is essential for developing successful investment strategies. To this end, the authors of this work propose a novel framework that predicts the price of Bitcoin (BTC), a dominant cryptocurrency. For stable prediction performance in unseen price range, the change point detection technique is employed. In particular, it is used to segment time-series data so that normalization can be separately conducted based on segmentation. In addition, on-chain data, the unique records listed on the blockchain that are inherent in cryptocurrencies, are collected and utilized as input variables to predict prices. Furthermore, this work proposes self-attention-based multiple long short-term memory (SAM-LSTM), which consists of multiple LSTM modules for on-chain variable groups and the attention mechanism, for the prediction model. Experiments with real-world BTC price data and various method setups have proven the proposed framework’s effectiveness in BTC price prediction. The results are promising, with the highest MAE, RMSE, MSE, and MAPE values of 0.3462, 0.5035, 0.2536, and 1.3251, respectively.

The authors of this work propose a deep learning-based fault detection model that can be implemented in the field of plastic injection molding. Compared to conventional approaches to fault detection in this domain, recent deep learning approaches prove useful for on-site problems involving complex underlying dynamics with a large number of variables. In addition, the advent of advanced sensors that generate data types in multiple modalities prompts the need for multimodal learning with deep neural networks to detect faults. This process is able to facilitate information from various modalities in an end-to-end learning fashion. The proposed deep learning-based approach opts for an early fusion scheme, in which the low-level feature representations of modalities are combined. A case study involving real-world data, obtained from a car parts company and related to a car window side molding process, validates that the proposed model outperforms late fusion methods and conventional models in solving the problem.