twumasiclement/TimeSeries-Forecasting

Machine Learning Algorithms for Forecasting and Backcasting Blood Demand Data with Missing Values and Outliers: A Study of Tema General Hospital of Ghana # Time-series prediction for blood demand (External R scripts and data for a paper submitted to the International Journal of Forecasting). Topic: Machine Learning Algorithms for Forecasting and Backcasting Blood Demand Data with Missing Values and Outliers: A Study of Tema General Hospital of Ghana. The 6 machine learning (ML) methods used were: K-Nearest Neighbor regression (KNN) Neural Network Auto-Regressive (NNAR) Generalized Regression Neural Network (GRNN) Multi-Layer Perceptron (MLP) neural networks Extreme learning machines (ELM) neural networks Long short-term memory (LSTM) neural network. # NB: The non-seasonal ARIMA model was considered as a baseline model for comparison with the ML models. These novel R functions (attached) were respectively developed to implement the six ML time-series models and the non-seasonal ARIMA model via a rolling-origin strategy for model comparison (for both forecast and backcast schemes). The main R script for the study is named `Main_Script_IJF.r` or `Main_Script_IJF.html` (i.e .R and .html file formats, respectively); and the other R scripts were external files sourced in the main script. The R codes were created with the help of other packages (properly refrenced in the main paper) and open-source R libraries Keras and TensorFlow. The empirical data on blood demand with missing values from Jan 2013 to Sept 2020 (which was used for model fitting) is also attached as `Blood_data_Tema.csv`. The external scripts (for implementing the rolling-origin forecast and backcast schemes) sourced in the main script `Main_Script_IJF.r` are: `rolling-eval-ARIMA-script` (ARIMA); `ro-eval-KNN-script.R` (KNN); `ro-eval-NNAR-script.R`(NNAR); `ro-eval-GRNN-script.R` (GRNN); `ro-eval-MLP-script.R` (MLP); `ro-eval-ELM-script.R` (ELM); `rolling-eval-LSTM-script.R`(LSTM). # Research gap and Contribution: Forecasters, policymakers, time-series users and practitioners usually need long time-series data for model assessment, policy analy …