In the rapid development of e-commerce today, traditional inventory management methods have encountered a large amount of inventory surplus and frequent out of stock problems. This article aims to combine machine learning with big data to optimize inventory of goods, in order to improve inventory efficiency and reduce costs. This article established a data collection platform to achieve real-time acquisition of sales data, inventory data, and market demand data, and to clean up the data to ensure its accuracy and completeness. On this basis, machine learning algorithms and statistical methods were used to monitor and analyze real-time data during the production process, timely detect abnormal situations such as sudden increase in sales and unsold goods, and initiate warnings. In response to abnormal situations and warning information, timely replenishment, adjustment of inventory and distribution strategies, and promotional activities were carried out to meet market demand and reduce inventory risks. This article used machine learning methods to predict sales and compared them with actual sales to analyze their prediction accuracy. The prediction for 4 items was correct; 1 item was incorrect; the prediction accuracy was 80%. This article was based on a machine learning historical data prediction method, which helped to monitor and manage inventory in real-time.