Comparison of the Performance of Multiple Linear Regression and Multi-Layer Perceptron Neural Network Algorithms in Predicting Drug Sales at Pharmacy XYZ

Arifuddin, Danang
Kusrini, Kusrini
Kusnawi, Kusnawi

Jurnal Sisfotek Global
Vol 15, No 1 (2025): JURNAL SISFOTEK GLOBAL

The needs of better drugs management tool especially that can predict specific drugs consumption volume are needed by any healthcare facility including retail pharmacies. Thus, finding better prediction algorithm with suitable variable internally and externally becoming this research objectives. The research compares correlation score and histogram of each predictor variable with target variable and further input the selected variable into MLR and MLPNN algorithm to find recommended algorithm with better MSE and MAPE. The findings indicate that MLPNN with backpropagation method slightly outperforms MLR with ‘h-7’ as single input variable but with unstable predictions with lower MSE of 19588 and MAPE of 22,3%. While MLR's MSE of 22346,129 and MAPE of 25.4% with ‘h-7’ and ‘bm’ as input variable perform stable prediction. Finally, the research find ‘h-7’ is the most significant variable among other variables and both MLR and MLPNN are both need better improvement to perform drugs prediction analysis.