Stable Decision Tree Method for Predicting Suicidal Ideation for At-Risk Homeless Youth

This project implements the stable decision tree algorithm based on the method outline in "Improving Stability in Decision Tree Models"[^1] that presentas a unique distace metric for heuritic-based decision trees as a measure of stability. The algorithm produces a Pareto optimal set from which a single final optimal tree is selected according to an objective function targeting a unique metric to optimize (AUC, distance, combined, etc.). Our Implementation attempts to improve upon previous work[^2] in creating an effective method to identify suicide risk among youth experiencing homelessness(YEH). The dataset used in this implementation presents a unique contribution to considering social network features as well as the individual factors in building risk profiles.

The distance metric implementation used in the code may be found as a reference below.[^3]

We reproduce the original aggregation metrics from the Bertimas paper, comparing different Pareto optimal tree selection strategies and a downsampled variant. Results can be found in the Experiments folder

A write-up of the project and experiment design can be found here: docs/Bertimas-Report-Final.pdf

[^1]: Improving Stability in Decision Tree Models

[^2]:"Getting to the Root of the Problem: A Decision-Tree Analysis for Suicide Risk Among Young People Experiencing Homelessness"

[^3]: Path Distance Metric Repository from Stable Decision Tree Algorithm

Commands to run

```bash uv run src/StableTree/main.py --group-name FINALaggregateoutput --option experiment --datasets data/DataSetCombinedSISNIBaselineFE.csv data/DataSetCombinedSISNIBaselineFE.csv data/breast_cancer.csv --labels suicidea suicattempt target

uv run src/StableTree/main.py --group-name finalaggregateoutputalldatasets --option plot --datasets data/DataSetCombinedSISNIBaselineFE.csv data/breastcancer.csv
```

Setup python & env

1. install uv curl -LsSf https://astral.sh/uv/install.sh | sh 33 brew install graphviz #graphviz binaries for pydotplus

2. cd to source directory; file(s) using UV cd suicide_project uv venv # only first time source /bin/activate uv sync uv run run8.py uv run run9.py

## Terminal Example:

Running for dataset DataSetCombinedSISNIBaseline_FE with seed 42

================================================== dsnameDataSetCombinedSISNIBaselineFE.csv

Experiment: experiment20250501134848seed42DataSetCombinedSISNIBaselineFEsuicidea - Seed: 42 - Dataset: DataSetCombinedSISNIBaselineFE

Number of samples in the full dataset: 586

Number of samples in the training set: 726

Number of samples in the test set: 242

Shape of training set: (726, 56)

Shape of random split: (363, 56), (363,)

Number of trees in T0: 20

Number of trees in T: 20

Computing average tree distances || 20/20 [100%] in 20.7s (0.96/s)

Number of distances computed: 20

Average AUC score: 0.821854723038044

Number of Pareto optimal trees: 7

Frequenicies of top 2 common features: [[('traumasum', 70.0), ('fight', 20.0)], [('harddruglife', 45.0), ('exchange', 15.0)], [('LEAFNODE', 25.0), ('harddruglife', 20.0)]]

Selected stability-accuracy trade-off final tree index: 1

Stability-accuracy tree depth: 4, nodes: 23

Selected AUC maximizing tree index: 1

AUC-maximizing tree depth: 4, nodes: 23

Selected distance minimizing tree index: 15

Distance-minimizing tree depth: 11, nodes: 79

Completed experiment: experiment20250501134848seed42DataSetCombinedSISNIBaselineFE_suicidea

References: