Generative Machine Learning model trained on UD Police Daily Statistics from 2017-2021. When provided a LOCATION, DATE, & TIME it generates a prediction of what crime description may be committed.
Web scraping Data
- First, the data needed to be gathered by the UD Police Statistics website. It was not easily available to download and was separated by day.
- A simple Python script got all the data into a CSV file.
Cleaning Data
- Although the data seemed usable it required some standardizing for an ML model.
- The entries contained human error and typos so there was a lot of duplicate data with incorrect spelling or formatting. For example:
Trabant Student Center&Trabant Building - The dates needed to be separated into DAY, MONTH, YEAR
- The times needed to be standardized to military time and
:removed
Training the model
- Using
sklearnthe model was trained on a DecisionTreeClassifier() - The data was split into training and testing groups. The test size was 20% of the data.
- All the data was encoded so that they were numerical values because ML models are essentially mathematical models.
- The range of accuracy (since March 2024) is 20% - 30%
Getting Predictions
# Predict Crime Description for given Location and date/time
sample_location = "Smith Hall"
sample_time = 1200
sample_day = 3
sample_month = 5
sample_year = 2024
sample_location_encoded = label_encoders["Location"].transform([sample_location])[0]
predict_description = clf.predict([[sample_time, sample_location_encoded, sample_day, sample_month, sample_year]])
# Print the inverse encoding (readable text)
print(label_encoders["Description"].inverse_transform(predict_description))
Output: ['Trespass']