Business Decision-Making Using Hybrid LSTM for Enhanced Operational Efficiency

In the evolving landscape of the modern business ecosystem, characterized by rapid globalization and the emergence of Industry 4.0 technologies, there is a heightened demand for intelligent decision-making across diverse sectors ranging from production and finance to human resources. A pivotal component driving these decisions is Sales Forecasting (SF), which influences consequential facets like production optimization and inventory management. While various models exist for SF, the long short-term memory (LSTM) model, despite its effectiveness, presents certain challenges. This paper introduces a novel hybrid model that seamlessly integrates a bidirectional long short-term memory network (BiLSTM) with an advanced embedding and dense framework, aiming to address the recognized constraints of standard LSTM. The core architecture employs an embedding layer, transforming sales data into a multidimensional matrix, followed by a series of interconnected BiLSTM layers equipped with skip connections. The final segments of the architecture consist of dense layers, resulting in a refined SF. Upon experimentation with an industrial dataset, preliminary results indicate that our model surpasses the performance of traditional CNN, GRU, and LSTM models, marking a significant stride in SF methodologies.