In continuation with last week’s post, today I will be writing about the second article that I included in my research paper. Although very similar to the previous article, today’s article is about the applications of machine learning in identifying young offenders. In the experiment, which was done in 2023, Bonfante and her team analyzed the underlying factors that could be used to predict aggresive behavior in young individuals. Through experiments done on the delinquent and non-delinquent youths, the researchers extracted 37 predictor variables. After acquiring these variables, the researchers used three machine learning algorithms to see which one was the most effective in distinguishing between a sex offender and a non-sex offender. Today, I will be going over the three machine learning algorithms and explaining each one’s significance in the experiment.
The first machine learning algorithm used in the experiment was Support Vector Machine (SVM). The SVM algorithm was crucial for analyzing cognitive function data that differentiated offenders and non-offenders. It was important because it effectively handled large datasets and found the optimal boundary between possible outcomes, making it a robust choice for classification tasks. Its significance to the experiment lies in its ability to identify complex patterns within the data, which can indicate underlying cognitive deficits associated with antisocial and aggressive behaviors. The experiment revealed that while the K-NN model performed best, SVM still showed notable efficacy, especially when trained with features selected by the filter method. Put simply, the filter method involved selecting relevant data for the experiment, helping to sort through data points; keeping only those that were really important for making predictions. SVM’s accuracy in the experiment demonstrates its relevance in predicting behaviors based on cognitive assessments, thereby influencing the experiment’s outcomes and contributing to the understanding of cognitive functions concerning criminal behavior.
The second machine learning method was Random Forest, an algorithm used for classification tasks. It works by growing a collection of decision trees and then outputting the average final decision. It was beneficial because it also handled large amounts of data with high accuracy. Random Forest was important to the experiment because it helped in feature selection. The Boruta algorithm used in the experiment was built around Random Forest and was used to determine the most important features of the data. By identifying these key features, the Random Forest model contributed significantly to refining the dataset, making it more manageable and improving the overall predictive accuracy of the machine learning models.
The third and final machine learning model used was K-Nearest Neighbors (K-NN). K-NN is an algorithm used to predict the classification or value of a given data point. It first chooses the number of data points it will look at (these are called the neighbors), then measures the distance between the data points that you want to classify and finds the closest neighbors to your data point. Following that, it begins to classify by looking at the categories of the neighbors and choosing the most common one. The purpose of K-NN is to predict what category a set of data belongs to by analyzing the other “neighbors” and generating categories. In the Bonfante experiment, K-NN was crucial because it provided the highest balance accuracy, meaning it was able to classify the events and data given to it with the most effectiveness. It was superior to any other machine learning model because it classified young offenders from non-offenders with the most accuracy. More specifically, accuracy refers to the machine learning’s ability to be efficient in utilizing variables. During the experiment, the researchers had 37 different features they could use to make their predictions, but using all 37 was not as useful as just picking the most helpful out of the 37. Hence, why it was the most successful since the K-NN model only used 19 of the features to make the best prediction, while the SVM model had to use 24 variables.
In summary, this experiment demonstrates the vast potential of machine learning in cognitive behavior. The findings contribute to a deeper understanding of the cognitive factors associated with antisocial behavior and suggest ways we can prevent it. Althouhg, further research with larger sample sizes and the exploration of new technologies would be able to expand these findings even more, the article exhibits the important steps researchers are taking to utilizing machine learning for the benefit of others.
Data Perspectives 
Leave a comment