Sales Prediction of Kembar Fruit Salad Homemade Products Based on Transaction Data Using K-Nearest Neighbor

Abstract

K-Nearest Neighbor is a method used to classify data with the closest distance or what is called a lazy learning technique. Accurate sales predictions allow owners to plan raw material stock requirements more efficiently, thereby reducing the risk of overstocking (which can cause waste) or understocking (which has the potential to reduce revenue). This research will implement the KNN, K-Nearest Neighbor algorithm to show the extent to which the KNN method can produce accurate predictions in the context of sales. The stages of the method carried out in this study by determining the value of K, calculating the square of the euclid distance (query instance) of each object against the given training data, then sorting the objects into groups that have the smallest euclid distance, using the class label Y (nearest neighbor classification), using the k-nearest neighbor category that is the majority then the calculated query instance value can be predicted. This research produces a jupyter notebook-based sales prediction model that can be used to predict Twin Fruit Salad products based on variations. The dataset used consists of 112 data divided into 89 training data and 23 testing data. With 91% accuracy results, it contributes to increasing the turnover of twin fruit salad and can provide considerations regarding stock availability based on the amount sold.