Menu
close

Risk prediction with AI to optimise resource allocation

Artificial Intelligence is well known for its content generation capabilities, but it can also be used for other purposes that have a positive social impact, such as predicting high-risk crime areas.

Thanks to the study of historical data and socio-economic, climatic or temporal variables, it is possible to predict where there is a higher risk of crime, allowing the optimisation of police resources and understanding the dynamics of crime, in this case in the city of Montevideo.

Carried out by Marcelo Miraballes | Daniel Vega | Diego Macias | Marco Antonio Peña

Qualification Master in IA & Data Science

Technologies Artificial Intelligence | Machine Learning | Data Analysis and Visualisation | Optimisation of Models | Development Environments

What is the motivation?

The use of AI is widespread in fields such as content generation, but it also offers unprecedented opportunities for social advances. The rise of data collection has increased interest in applying Machine Learning techniques in the service of public safety, analysing data to predict areas of high crime risk and facilitating efficient resource allocation.

Program aims

  • Predict high-risk crime areas using historical data and socio-economic, climatic and temporal variables.
  • Optimising patrol resources through a data-driven approach.
  • Analysing crime patterns to better understand the dynamics of crime in Montevideo.

Development

The project has involved the following phases:

  • Collection of historical data and different climatic and socio-economic variables.
  • Data cleaning and preparation, adjusting variables and dealing with missing values.
  • Exploratory data analysis, identifying different crime patterns, analysing correlations and detecting outliers.
  • Feature engineering, analysing the importance of different variables, encoding categorical variables and creating new variables.
  • Predictive modelling, training the different supervised models, optimising the hyperparameters and carrying out an evaluation with performance metrics.
  • Evaluation of the model, studying the ROC curve and overfitting, optimising the classification threshold and evaluating the metrics.

Results

The results obtained include:

  • Robust model with an AUC above 0.75 and an F1-Score optimised by threshold adjustment techniques.
  • Optimised model performance with the use of Optuna and efficient hyperparameter tuning.

Conclusions

The rise of artificial intelligence has focused on its generative characteristics, but it can also be used to improve society. AI- and ML-supported crime analysis in areas critical to public safety facilitates the optimisation of police resources, allocating them where they are most needed, as well as supporting urban planning and crime prevention of all kinds.

arrow-right