Machine Learning Based Office Stationery Classification (ATK) Analysis Using EDA

Abstract

The need for efficient office stationery (ATK) management is increasing along with the development of the business world and the digitalization of business processes. Manual grouping of ATKs tends to be time-consuming, inconsistent, and error-prone. Therefore, this research aims to develop a machine learning-based ATK classification system to support decision making in inventory management. The research was conducted at the Martin Photocopy Shop, with a quantitative approach using ATK sales data over a certain period. The research stages include data collection, data cleaning, exploratory analysis using Exploratory Data Analysis (EDA) techniques, as well as the application of two machine learning algorithms, namely the Random Forest Classifier and Support Vector Machine (SVM). Data labeling is carried out based on the categories of Best-Selling, Less-Selling and Non-Selling products. The results show that the SVM algorithm provides the best performance with accuracy of 85%, precision 0.79, recall 0.67, and F1-score 0.88, ahead of Random Forest which recorded accuracy of 80% and F1-score 0.85. EDA also succeeded in uncovering important patterns in the data, such as the positive relationship between stocks and sales quantities. These findings confirm that the integration of machine learning and EDA is able to identify product characteristics more accurately, as well as provide strategic insight for stock management. In conclusion, the use of machine learning-based classification methods is very effective in supporting store operational efficiency, avoiding excess or shortage of stock, and improving service to customers.