Open AccessA Sales Prediction Method Based on LSTM with Hyper-Parameter Search
Author(s) -
Yun Dai,
Jinghao Huang
Publication year - 2021
Publication title -
journal of physics. conference series
Language(s) - English
Resource type - Journals
SCImago Journal Rank - 0.21
H-Index - 85
eISSN - 1742-6596
pISSN - 1742-6588
DOI - 10.1088/1742-6596/1756/1/012015
Subject(s) - computer science , machine learning , artificial intelligence , field (mathematics) , face (sociological concept) , series (stratigraphy) , time series , function (biology) , preference , regression , data mining , statistics , mathematics , paleontology , social science , evolutionary biology , sociology , pure mathematics , biology
Sales forecast is a significant topic in business operation, which generally formulated as a time-series regression problem. Although there are many research results in this field, we still face some challenges in real scenes, such as data with high-sparsity, users may have a preference in prediction results, and systems need a single model with high performance. In this paper, a method is proposed to address the above challenges. We present a long short-time memory (LSTM) model with a special loss function and use the hyper-parameter search for accuracy optimization. To illustrate the performance, we employ them on the open dataset, Kaggle Rossman sales data. The experiment results show that compare with a series of machine learning models using the AutoML (Auto Machine Learning) tool, the proposed method significantly increased the performance of prediction on sparse data. Besides, it can reasonably overestimate or underestimate sales forecasts based on user preferences that meet the actual business demands.