Predictive Data Science Models
Date of request: 3 June 2025
Reference: 698-25
Request
The information I require is as follows and relates to ‘predictive’ data science models being used by Avon and Somerset Police. Please may you provide me with the following:
1. As of June 1, 2025, a list of ‘predictive’ data science models being used by Avon and Somerset Police
2. For each of the models in response to question 1, since January 1, 2025: model Accuracy, Precision, Recall, Matthews Correlation Coefficient scores produced by validation and audit tests
3. For each of the models, since January 1, 2025: any audit/evaluation data related to true and false positives (and similar metrics)
4. For each of the models, a list of indicators/data points that are used by the models
Response
Questions 1 and 2
| Model | Accuracy | Precision | Recall | MCC | |
| 1 | CV_Victim_Violence_Against_The_Person | 0.991214369 | 0.953917051 | 0.943052392 | 0.949650839 |
| 2 | CO_Offender_Violence_Against_The_Person | 0.970519328 | 0.725490196 | 0.375634518 | 0.719912507 |
| 3 | AV_Victim_Violence_Against_The_Person | 0.999541074 | 0.992716367 | 0.998706897 | 0.992304206 |
| 4 | AO_Offender_Violence_Against_The_Person | 0.999242772 | 0.984714874 | 0.995838288 | 0.984094037 |
| 5 | Victim | 0.994370305 | 0.903600832 | 0.62815333 | 0.902575181 |
| 6 | Offender | 0.992479443 | 0.755047293 | 0.339120036 | 0.711620906 |
| 7 | Non_Crime_Level_1 | 0.918799952 | 0.761111111 | 0.251645492 | 0.753099186 |
| 8 | Non_Crime_Level_2 | 0.64262965 | 0.708017861 | 0.140642474 | 0.683527932 |
| 9 | Non_Crime_Level_3 | 0.594511 | 0.565957 | 0.894428 | 0.730193 |
For completeness, descriptions for each model are provided below:
CV_Victim_Violence_Against_The_Person
Risk of being linked as a repeat child victim of Violence against the person
CO_Offender_Violence_Against_The_Person
Risk of reoffending as a child of violence against the person
AV_Victim_Violence_Against_The_Person
Risk of being linked as a repeat adult victim of Violence against the person
AO_Offender_Violence_Against_The_Person
Risk of reoffending as an adult of Violence against the person
Victim
Risk of being linked to a repeat vulnerable incident (victim of a crime or missing person etc)
Offender
Risk of reoffending
Non_Crime_Level_1
Data Quality model to find Level one crimes that have been categorised as a Non-Crime that contain a crime
Non_Crime_Level_2
Data Quality model to find Level two crimes that have been categorised as a Non-Crime that contain a crime
Non_Crime_Level_3
Data Quality model to find Level three crimes that have been categorised as a Non-Crime that contain a crime
Question 3
This is not recorded information.
The processed data in relation to true and false positives and similar metrics is recorded as model scores.
Question 4
For Models 1 – 6 listed above, please find the requested information in the attached file titled, ‘Variables List’.
For Models 7 – 9 listed above, please find the requested information in the attached file titled, ‘Word List’.