Ana Billah, Aulia Yolanda, Dea Khairani, Dimas Triono Putra, Helma Tiara Syif

Prediksi Jumlah Obat Menggunakan Jaringan Syaraf Tiruan RNN pada Data Penjualan Bulan Juli

Introduction

Prediksi jumlah obat menggunakan jaringan syaraf tiruan rnn pada data penjualan bulan juli . Prediksi jumlah obat dengan Jaringan Syaraf Tiruan RNN untuk manajemen inventori apotek/rumah sakit. Studi ini membandingkan kinerja RNN dan LSTM pada data penjualan Juli.

Abstract

Accurate drug quantity predictions are crucial in inventory management at pharmacies or hospitals to ensure sufficientdrug availability and avoid overstocking or stockouts. However, these predictions are often difficult to make due tocomplex and dynamic drug sales patterns. This study aims to predict drug sales volume using Recurrent Neural Network(RNN) and Long Short-Term Memory (LSTM). The dataset was collected from pharmacy sales records in July. Theresearch stages included data preprocessing, normalization, constructing a time series dataset with a window size of 3,and splitting into training (80%) and testing (20%) datasets. The models were trained for 100 epochs with a batch size of10. The results show that the RNN model achieved a Root Mean Squared Error (RMSE) of 338.16, while the LSTM modelproduced an RMSE of 433.44. This indicates that RNN outperformed LSTM in predicting drug sales on a simple dataset.The findings suggest that RNN can serve as an alternative method to support drug stock planning to ensure betterdistribution and availability

Review

This paper addresses a highly relevant and critical problem in healthcare inventory management: accurate drug quantity prediction. The study leverages Recurrent Neural Networks (RNN) and Long Short-Term Memory (LSTM) to forecast drug sales volume, a crucial endeavor for ensuring drug availability while minimizing waste. The core finding indicates that the RNN model achieved superior performance over LSTM, with a lower Root Mean Squared Error (RMSE), suggesting its potential as a viable alternative for optimizing drug stock planning, particularly within the context of pharmacy sales data for the month of July. Methodologically, the paper outlines a clear and structured approach, commencing with essential data preprocessing and normalization steps. The construction of a time series dataset with a window size of 3 and a standard 80/20 train-test split demonstrates sound practice in preparing data for neural network training. Furthermore, the explicit reporting of training parameters, such as 100 epochs and a batch size of 10, enhances reproducibility. The comparative analysis between RNN and LSTM, backed by specific RMSE values (338.16 for RNN and 433.44 for LSTM), provides concrete evidence for the models' performance and the relative effectiveness of RNN in this particular application. While the initial findings are promising, a significant limitation lies in the scope of the dataset, which is restricted to "Data Penjualan Bulan Juli" (July sales data) and described as a "simple dataset." This limited temporal and potentially contextual scope raises concerns about the generalizability of the results to longer periods, different seasons, or more complex sales patterns typically encountered in real-world scenarios. Future work would greatly benefit from expanding the dataset to encompass a full year or multiple years of sales data, possibly from several pharmacies, to validate the models' robustness and effectiveness across diverse conditions. Additionally, further discussion on the practical significance of the reported RMSE values relative to typical drug sales volumes, and exploration of other performance metrics or statistical comparisons between models, would enrich the analysis.

Full Text

Comments

You need to be logged in to post a comment.