Accurate sales forecasting is critical for optimizing supply chain decisions, particularly in a business environment where unpredictable and sparse occurrences characterize intermittent demand data. Intermittent type of data, typical in customer demand or sales data for items with irregular sales patterns, presents unique challenges due to its nonlinear nature and limited dataset volume. Traditional forecasting approaches often fail to adequately address these challenges, leading to suboptimal decision-making. This paper evaluates the application of advanced neural network models, specifically Long Short-Term Memory (LSTM) and Multilayer Perceptron (MLP), to improve forecasting accuracy for intermittent demand. Croston’s method is commonly used to predict intermittent demand data. We adapt these models to effectively capture the stochastic behaviours of the demand data, increasing their predictive power under conditions of irregular demand. We have applied our models to an extensive dataset spanning five years, including 313 unique products and 244 locations (pin codes) over the same period. Approximately 55% of the data entries showed zero demand for specific items at various locations, making it an ideal dataset for forecasting intermittent demand patterns.The results indicate that the proposed method improves forecasting accuracy compared to various machine learning models. This study highlights the potential of the LSTM model to deliver more precise sales forecasting, which in turn can lead to better supply chain management.