NCJ Number
45797
Journal
ENERGY AND TECHNOLOGY REVIEW Dated: (NOVEMBER 1976) Pages: 5-10
Date Published
1976
Length
6 pages
Annotation
THE DEVELOPMENT OF AN AUTOMATED CRIME ANALYSIS CAPABILITY, WHICH WILL EVENTUALLY BE OPERATIONALIZED TO AID THE SAN DIEGO (CALIF.) POLICE DEPARTMENT IN THE EFFECTIVE DEPLOYMENT OF MANPOWER, IS DESCRIBED.
Abstract
THE FIRST PHASE OF THE TECHNOLOGY TRANSFER PROJECT INVOLVED ASSESSING THE FEASIBILITY OF APPLYING COMPUTER PATTERN RECOGNITION TECHNIQUES TO ACTUAL CRIME STATISTICS WITH THE AIM OF PREDICTING THE LIKELIHOOD OF A CRIME BEING SOLVED BASED ON SUCH VARIABLES AS TYPE OF CRIME, LOCATION, AND TIME OF CRIME. A REVIEW OF 11,645 CASE RECORDS RESULTED IN A SUMMARY OF STATISTICALLY DIFFERENT CRIMES RANGING FROM PETTY THEFT TO HOMICIDE. IN INITIAL EXPERIMENTS, PATTER WAS USED TO EVALUATE THE EFFICIENCY OF VARIOUS TECHNIQUES IN PREDICTING THE SUSCEPTIBILITY TO SOLUTION OF NEWLY REPORTED CRIMES. PATTER IS A COMPUTER PROGRAM CODED IN FORTRAN IV WHICH IS SPECIFICALLY SUITED FOR MODELING ANALYTIC PROCESSES AND WHICH CAN BE USED IN AN INTERACTIVE MODE. FOR A GIVEN COLLECTION OF DATA, THE SYSTEM WILL TRY TO DETERMINE IF THE PREDICTION OF AN UNMEASURED PROPERTY OF THE DATA IS POSSIBLE BASED UPON THE INFORMATION AT HAND. EXPERIMENTS BASED ON FOUR BASIC VARIABLES PROVIDED INSUFFICIENT INFORMATION FOR ACCURATE PREDICTION. AN ANALYSIS OF 100 CRIME CASES USING 7 VARIABLES ACHIEVED A 15 PERCENT BETTER PREDICTIVE ACCURACY. BASED ON THESE RESULTS IT WAS DECIDED THAT PATTERN RECOGNITION WAS FEASIBLE FOR CRIME ANALYSIS. PHASE 2 INVOLVED SELECTING AN OPERATIONALLY REASONABLE MODEL AND OPTIMIZING THE PATTERN RECOGNITION PERFORMANCE OF THE AVAILABLE VARIABLES WITHIN THE CONTEXT OF THE MODEL. FURTHER DATA ANALYSIS RESULTED IN THE IDENTIFICATION OF 10 VARIABLES: EAST/WEST COORDINATES, NORTH/SOUTH COORDINATES, THE CROSS PRODUCT OF BOTH SETS OF COORDINATES, HOUR OF CRIME OCCURRENCE, DAY OF WEEK OF OCCURRENCE, TIME OF CERTAINTY (I.E. DAYLIGHT, NIGHT), TYPE OF CRIME, CLOSURE RATE, OCCURRENCE RATE, AND THE PRODUCT OF CLOSURE AND OCCURRENCE RATES. BASED ON A TRAINING SET OF 500 CRIMES, CLASSIFICATION RULES WERE DEVELOPED. TO TEST THE ACCURACY OF THE PREDICTIVE RULES, A TEST DATA SET OF 200 CRIMES WAS PREPARED. TEST CASES WERE RANKED IN ORDER OF PROBABILITY OF SOLUTION. ACCORDING TO THE RECORDS, 34 OF THE 200 CASES HAD BEEN SOLVED. OF THESE SOLVED CASES, PATTER HAD LISTED 20 AMONG THE FIRST 76 OF THE LISTING. THE RELATIVE COST EFFECTIVENESS OF SOLVING 20 OUT OF 76 (26 PERCENT SUCCESS RATE) BY AUTOMATED CRIME ANALYSIS, AS COMPARED WITH THE 17 PERCENT ACTUAL SUCCESS RATE IS ENCOURAGING. AS PART OF PHASE 3, ADDITIONAL VARIABLES WILL BE INCLUDED FOR ANALYSIS IN THE PATTERN RECOGNITION PROCESS, AND IT WILL BE DECIDED WHETHER THESE ADDITIONAL VARIABLES SHOULD ALSO BE INCLUDED IN THE DATA FIELDS OF THE AUTOMATED REGIONAL JUSTICE INFORMATION SYSTEM. FINALLY THE MORE EFFICIENT ALGORITHMS WILL BE SELECTED FOR INCORPORATION IN THE SAN DIEGO SYSTEM AND APPROPRIATE PATTERN RECOGNITION TECHNIQUES WILL BE IMPLEMENTED.