In the field of data mining the high-utility itemset mining (HUIM) is a necessary machine that allows it to search for itemsets in transactional data which are useful. However, Traditional HUIM methods commonly fail in scalability and efficiency because of the combinatorial explosion of candidate itemsets. In this research, a original method to achieve HUIM is proposed by combining Deep Learning (DL) techniques to increase both the accuracy and efficiency. In this case, we proposed a Deep Learning technique based on a neural network architecture that quantifies and dynamically extracts HUIs from very large transactional datasets. In particular, we uses a Long Short Term Memory (LSTM) networks to record sequential dependencies in the transactions, and Convolutional Neural Networks (CNNs) to extract features. We integrate an Ant-Colony-Optimization algorithm to optimize the DL approaches’ hyperparameters and the feature selection procedure to further improve additional development performance. Evaluation on benchmark transactional datasets revealed significant improvements in mining accuracy, processing time and overall utility discovery of the proposed technique comparing with traditional HUIM approaches. The results of this research suggest a solid framework for application of DL into HUIM, enabling more efficient data analysis and decision making in a variety of areas.