https://github.com/nrennie/outlier-detection-in-network-revenue-management
R Code corresponding to the paper "Outlier detection in network revenue management".
https://github.com/nrennie/outlier-detection-in-network-revenue-management
publication-code
Last synced: about 2 months ago
JSON representation
R Code corresponding to the paper "Outlier detection in network revenue management".
- Host: GitHub
- URL: https://github.com/nrennie/outlier-detection-in-network-revenue-management
- Owner: nrennie
- License: cc-by-4.0
- Created: 2021-03-08T16:18:15.000Z (about 4 years ago)
- Default Branch: main
- Last Pushed: 2024-01-31T13:19:16.000Z (over 1 year ago)
- Last Synced: 2025-02-08T08:23:01.021Z (4 months ago)
- Topics: publication-code
- Language: R
- Homepage:
- Size: 277 KB
- Stars: 2
- Watchers: 1
- Forks: 0
- Open Issues: 0
-
Metadata Files:
- Readme: README.md
- License: LICENSE
Awesome Lists containing this project
README
# Outlier detection in network revenue management
Code corresponding to the paper "Outlier detection in network revenue management". Previously submitted to arXiv as ["Detecting outlying demand in multi-leg bookings for transportation networks"](https://arxiv.org/abs/2104.04157).
This code is written using [R](https://www.r-project.org/) which can be installed from [cran.r-project.org/bin/windows/base](https://cran.r-project.org/bin/windows/base/). The following R packages are also required for this analysis:
* POT
* tidyverse
* mrfDepth
* fdapace
* MASS
* igraph
* forecast
After installation, these can be loaded using the `required_packages.R` script.
## Prepare the data
### `extrapolation_function.r`
Forecasts the remaining bookings for those booking patterns which have not yet departed. The historic data forecasts include day of departure as a factor. Calls `historic_forecast_function.R`.
### `residuals_function.r`
Applies a functional regression to calculate the residual booking patterns. The default is to include only weekday of departure in the model. You can easily add in factors for month, year of departure etc.
## Determine the clusters
### `correlation_matrix_function.R`
Takes as input a list of legs for which the correlations should be calculated. Returns a matrix with containing the functional dynamical correlations.
### `mst_clustering_threshold.R`
Returns a list of clusters where each list item contains a vector of leg names in each cluster. This function calls `invert_graph.R`.
## Find and aggregate the outliers (run for each cluster)
### `depth.R`
This function takes the output of residuals function as input (plus optional agruments). This should be run for each leg.
This function calls `depth_threshold.R`. The output is a named vector (names correspond to a uniqueID i.e. departure date).
### `merge_differences.R`
Input is a named list of vectors. Each vector is the output of `depth.R` and there should be one vector for each leg within the cluster.
### `gpd_probs.R`
Takes the output of `merge_differences.R` as input and produces a data frame of ranked outliers with columns (unique ID, outlier probability, legs within cluster detected in).