Data Mining Project: Forecasting In-Patient & Day Case Waiting Time

The aim of this project is to forecast the number of patients in each time-band. We made use of the historical data provided by The National Treatment Purchase Fund.

The patients in the waiting list are grouped in one of these categories:

• 0-3 months
• 3-6 months
• 6-9 months
• 9-12 months
• 12-15 months
• 15-18 months
• 18+ months

Models implemented in this project:

• ARIMA (AutoRegressive Integrated Moving Average)
• SARIMA (Seasonal AutoRegressive Integrated Moving Average)
• TBATS (Trigonometric seasonality, Box-Cox transformation, ARMA errors, Trend and Seasonal components)
• Facebook/Meta Prophet

The results obtained from running those models can be observed in the report as well as the .ipynb files (runnable in Google Colab).

Jupyter Notebooks

1. merging_data_2014_2021.ipynb

   • Dataset columns renamed: column headers between 2017 to 2021 were unaligned and needed cleaning to be done.
   • The separate .csv files are merged into one .csv file.
   • The merged .csv file is named merged.csv which can be found in the root folder of the repo.

2. eda_and_modelling.ipynb

   • Loaded in merged dataset: data between 2014 and 2021 (March)
   • Checked for missing data
   • Extracted out the relevant data from the dataset - column count
   • Aggregated the data per month
   • Basic data visualisation - moving average, exponential smoothing, ETSD (Error Trend Seasonality Decomposition)
   • ADF (Augmented Dickey-Fuller) statistical test
   • Split dataset into train and test sets
   • Implementation of ARIMA - plotted and evaluated
   • Implementation of SARIMA - plotted and evaluated
   • Implementation of TBATS - plotted and evaluated
   • Implementation of Prophet - plotted and evaluated