GENETIC NEURAL NETWORK BASED DATA MINING AND APPLICATION IN CASE ANALYSIS OF POLICE OFFICE



Similar documents
A Study of Discovering Customer Value for CRM:Integrating Customer Lifetime Value Analysis and Data Mining Techniques

Inventory Management MILP Modeling for Tank Farm Systems

A multi-item production lot size inventory model with cycle dependent parameters

12/7/2011. Procedures to be Covered. Time Series Analysis Using Statgraphics Centurion. Time Series Analysis. Example #1 U.S.

TECNICHE DI DIAGNOSI AUTOMATICA DEI GUASTI. Silvio Simani References

The Prediction Algorithm Based on Fuzzy Logic Using Time Series Data Mining Method

Spline. Computer Graphics. B-splines. B-Splines (for basis splines) Generating a curve. Basis Functions. Lecture 14 Curves and Surfaces II

The Virtual Machine Resource Allocation based on Service Features in Cloud Computing Environment

Forecasting the Demand of Emergency Supplies: Based on the CBR Theory and BP Neural Network

GUIDANCE STATEMENT ON CALCULATION METHODOLOGY

An Ensemble Data Mining and FLANN Combining Short-term Load Forecasting System for Abnormal Days

How To Calculate Backup From A Backup From An Oal To A Daa

HEAT CONDUCTION PROBLEM IN A TWO-LAYERED HOLLOW CYLINDER BY USING THE GREEN S FUNCTION METHOD

Projective geometry- 2D. Homogeneous coordinates x1, x2,

Genetic Algorithm with Range Selection Mechanism for Dynamic Multiservice Load Balancing in Cloud-Based Multimedia System

Capacity Planning. Operations Planning

MODEL-BASED APPROACH TO CHARACTERIZATION OF DIFFUSION PROCESSES VIA DISTRIBUTED CONTROL OF ACTUATED SENSOR NETWORKS

A Hybrid AANN-KPCA Approach to Sensor Data Validation

Proceedings of the 2008 Winter Simulation Conference S. J. Mason, R. R. Hill, L. Mönch, O. Rose, T. Jefferson, J. W. Fowler eds.

A Hybrid Method for Forecasting Stock Market Trend Using Soft-Thresholding De-noise Model and SVM

Analyzing Energy Use with Decomposition Methods

PERFORMANCE ANALYSIS OF PARALLEL ALGORITHMS

MULTI-WORKDAY ERGONOMIC WORKFORCE SCHEDULING WITH DAYS OFF

Optimization of Nurse Scheduling Problem with a Two-Stage Mathematical Programming Model

An Anti-spam Filter Combination Framework for Text-and-Image s through Incremental Learning

Matrices in Computer Graphics

Selected Financial Formulae. Basic Time Value Formulae PV A FV A. FV Ad

MORE ON TVM, "SIX FUNCTIONS OF A DOLLAR", FINANCIAL MECHANICS. Copyright 2004, S. Malpezzi

Cooperative Distributed Scheduling for Storage Devices in Microgrids using Dynamic KKT Multipliers and Consensus Networks

Anomaly Detection in Network Traffic Using Selected Methods of Time Series Analysis

Methodology of the CBOE S&P 500 PutWrite Index (PUT SM ) (with supplemental information regarding the CBOE S&P 500 PutWrite T-W Index (PWT SM ))

INTERNATIONAL JOURNAL OF STRATEGIC MANAGEMENT

Management watch list $20.4B (326) and poorly performing

Lecture 40 Induction. Review Inductors Self-induction RL circuits Energy stored in a Magnetic Field

PARTICLE FILTER BASED VEHICLE TRACKING APPROACH WITH IMPROVED RESAMPLING STAGE

Linear Extension Cube Attack on Stream Ciphers Abstract: Keywords: 1. Introduction

Cost- and Energy-Aware Load Distribution Across Data Centers

Fundamental Analysis of Receivables and Bad Debt Reserves

The Rules of the Settlement Guarantee Fund. 1. These Rules, hereinafter referred to as "the Rules", define the procedures for the formation

Prioritized Heterogeneous Traffic-Oriented Congestion Control Protocol for WSNs

Index Mathematics Methodology

PerfCenter: A Methodology and Tool for Performance Analysis of Application Hosting Centers

Event Based Project Scheduling Using Optimized Ant Colony Algorithm Vidya Sagar Ponnam #1, Dr.N.Geethanjali #2

Pedro M. Castro Iiro Harjunkoski Ignacio E. Grossmann. Lisbon, Portugal Ladenburg, Germany Pittsburgh, USA

Efficiency of General Insurance in Malaysia Using Stochastic Frontier Analysis (SFA)

Currency Exchange Rate Forecasting from News Headlines

Forecasting Stock Prices using Sentiment Information in Annual Reports A Neural Network and Support Vector Regression Approach

DEPARTMENT OF ECONOMETRICS AND BUSINESS STATISTICS. Exponential Smoothing for Inventory Control: Means and Variances of Lead-Time Demand

Forecasting the Direction and Strength of Stock Market Movement

APPLICATION OF CHAOS THEORY TO ANALYSIS OF COMPUTER NETWORK TRAFFIC Liudvikas Kaklauskas, Leonidas Sakalauskas

Insurance. By Mark Dorfman, Alexander Kling, and Jochen Russ. Abstract

Nonlinearity or Structural Break? - Data Mining in Evolving Financial Data Sets from a Bayesian Model Combination Perspective

An Architecture to Support Distributed Data Mining Services in E-Commerce Environments

APPLYING LINGUISTIC PROMETHEE METHOD IN INVESTMENT PORTFOLIO DECISION-MAKING

UNIVERSITY TUITION SUBSIDIES AND STUDENT LOANS: A QUANTITATIVE ANALYSIS

Auto-tuning and Self-optimization of 3G and Beyond 3G Mobile Networks

SPC-based Inventory Control Policy to Improve Supply Chain Dynamics

Analysis of intelligent road network, paradigm shift and new applications

FRAMEWORK OF MEETING SCHEDULING IN COMPUTER SYSTEMS

What influences the growth of household debt?

Levy-Grant-Schemes in Vocational Education

Guidelines and Specification for the Construction and Maintenance of the. NASDAQ OMX Credit SEK Indexes

IMPROVING THE RESISTANCE OF A SERIES 60 VESSEL WITH A CFD CODE

A. Jagadeesan 1, T.Thillaikkarasi 2, Dr.K.Duraiswamy 3

Kalman filtering as a performance monitoring technique for a propensity scorecard

A Real-time Adaptive Traffic Monitoring Approach for Multimedia Content Delivery in Wireless Environment *

The Incentive Effects of Organizational Forms: Evidence from Florida s Non-Emergency Medicaid Transportation Programs

An Optimisation-based Approach for Integrated Water Resources Management

Prices of Credit Default Swaps and the Term Structure of Credit Risk

Ground rules. Guide to the calculation methods of the FTSE Actuaries UK Gilts Index Series v1.9

Market-Clearing Electricity Prices and Energy Uplift

Time Series. A thesis. Submitted to the. Edith Cowan University. Perth, Western Australia. David Sheung Chi Fung. In Fulfillment of the Requirements

COASTAL CAROLINA COMMUNITY COLLEGE

Preface. Frederick D. Wolf Director, Accounting and Financial Management Division

The Joint Cross Section of Stocks and Options *

The Sarbanes-Oxley Act and Small Public Companies

This research paper analyzes the impact of information technology (IT) in a healthcare

THE USE IN BANKS OF VALUE AT RISK METHOD IN MARKET RISK MANAGEMENT. Ioan TRENCA *

Social security, education, retirement and growth*

Multifunction Phased Array Radar Resource Management: Real-Time Scheduling Algorithm

Distribution Channel Strategy and Efficiency Performance of the Life insurance. Industry in Taiwan. Abstract

Nonparametric deconvolution of hormone time-series: A state-space approach *

Estimating intrinsic currency values

Omar Shatnawi. Eks p l o a t a c j a i Ni e z a w o d n o s c Ma in t e n a n c e a n d Reliability Vo l.16, No. 4,

Evolution with Individual and Social Learning in an. Agent-Based Stock Market

INTERNATIONAL EDUCATION, QUALIFICATION AND CERTIFICATION SYSTEMS IN WELDING. IIW International Institute of Welding

Infrastructure and Evolution in Division of Labour

Towards a Trustworthy and Controllable Peer- Server-Peer Media Streaming: An Analytical Study and An Industrial Perspective

Transcription:

GENETIC NEURAL NETWORK BASED DATA MINING AND APPLICATION IN CASE ANALYSIS OF POLICE OFFICE LIU Han-l, LI Ln, ZHU Ha-hong of Reource and Envronmen Scence, Wuhan Unvery, 9 Luoyu Road, Wuhan, P.R.Chna, 430079 Tel: 86-7-878809 E-mal: luhl000@ohu.com; lln@elecaro.com Keyword: Daa Mnng, Daa Warehoue, BP Neural Newor, Genec Algorm, Daa Cleanng, Cae Analy ABSTRACT: Th paper pu forward a meod a combne e learnng algorm of BP neural newor w genec algorm o ran BP newor and opmze e wegh value of e newor n a global cale. Th meod feaured a global opmzaon, hgh accuracy and fa convergence. The daa-mnng model baed on genec neural newor ha been wdely appled o e procedure of daa mnng on cae nformaon n e command cenre of polce offce. I acheve an excellen effec for ang people o olve cae and mae good decon. In paper, e prncple and meod of daa-mnng model are decrbed n deal. A real cae of applcaon alo preened. From cae we can draw a concluon a e daa-mnng model we have choen cenfc, effcen and praccable.. INTRODUCTION. The Defnon of Daa Mnng Daa mnng a procedure of exracon of nformaon and nowledge a are hded n daa, unnown by people and poenally ueful from a large quany of daa w mulple characerc a uncompleed, conanng noe, fuzzy and random. A a nd of cro-dcplne feld a yncreze mulple dcplne ncludng daabae echnology, arfcal nellgence, neural newor, ac, nowledge acquremen and nformaon exracon, nowaday daa mnng ha become one of e mo fron reearch drecon n e nernaonal realm of nformaon-baed decon mang. Analyng and comprehendng daa from dfferen apec, people ue daa mnng meod o dg ou ueful nowledge and hdden nformaon of predcon from a large amoun of daa a are ored n daabae and daa warehoue. The meod nclude aocaon rule, clafcaon nowledge, cluerng analy, endency and devaon analy a well a mlary analy. By fndng valuable nformaon from e analy reul, people can ue e nformaon o gude er bune acon and admnraon acon, or a er cenfc reearche. All of ee provde new opporune and challenge o e developmen of all nd of feld relaed o daa proceng.. The Applcaon of Daa Mnng Appled o e procedure of daa analyng, proceng and ulzaon a well a e procedure of decon mang n many ocal deparmen, daa warehoue and daa mnng echnologe a ee deparmen o mae cenfc and reaonable decon. Th ha profound meanng for e developmen of our ocey and economy. Daa mnng can be appled o varou dfferen realm. For nance, many ale deparmen ue daa mnng echnology o deermne e drbuon and e geographcal poon of e ale newor, e purchae and oc quane of every nd of good, n order o fnd ou e poenal cuomer group and adju e raege for ale. In nurance compane, oc compane, ban and cred card compane, people apply daa mnng echnology o deec e decepve acon of cuomer o reduce e commercal decepon. Daa mnng ha been alo wdely appled o medcal reamen and genec engneerng and many oer feld. In recen year, w e acceleraon of e ep of nformaon conrucon n polce deparmen and w e ncremen of developmen level, daa mnng echnology ha alo been appled o e polce deparmen epecally e command cenre of polce offce. Th paper manly dcue e prncple and e praccal applcaon of genec neural newor baed daa mnng model n cae analy of polce offce.. THE MEANING AND METHOD OF DATA MINING IN COMMAND CENTRE OF POLICE OFFICE. The Meanng of Daa Mnng n Command Cenre of Polce Offce Every day n e command cenre of polce offce, people receve a large number of nformaon abou cae receved w varou approache. The nformaon ha been npu no daabae o form a large amoun of cae nformaon. Thee cae nformaon ha been archved annually and perodcally o form a pleny of horcal cae reource. By nducng and analyng ee horcal cae, people can ge ome experence and learn ome leon a can help em olve cae and mae decon n e fuure. Therefore, n order o a polce deparmen o olve cae rapdly and mae decon effcenly, we hould yneze and organze ee horcal daa, ue proper daa mnng model o dcover e poenal and ueful nowledge behnd e daa, and en predc and analye e mporan facor n e daa ncludng e rae of crme, e conuon of crme populaon, e crme age rucure, e area drbuon of crme, e developng endency of crme, e mean and approache of crme, e hdden area of crmnal and o on. A preen all of ee have become urgen a a need our polce offce o accomplh n e procedure of daa proceng.. Two Sep of Daa Mnng The daa mnng procedure n e command cenre of polce offce manly nclude wo ep:

() Fr, flerng, elecng, cleanng and ynezng e archved horcal cae nformaon, and en performng ranformaon f neceary, fnally, loadng daa no daa warehoue afer above proceng. () Choong approprae model and algorm of daa mnng o dg ou e poenal nowledge n daa. By pleny of analy and comparon among varou daa mnng model, we elec e error bac propagaon (BP) neural newor a e general-purpoe calculaon model n our daa mnng. We ran e neural newor w a uperved learnng meod and combne BP algorm w genec algorm o opmze e value of wegh. Furer, we apply e raned model o e predcon, clafcaon and rule exracon of e cae nformaon. 3. DATA MINING MODEL OF NEURAL NETWORK 3. Common Meod of Daa Mnng Preen daa mnng meod nclude ac meod, aocaon dcovery, cluerng analy, clafcaon and regreon, OLAP(On Lne Analycal Proceng), query ool, EIS(Execuve Informaon Syem), neural newor, genec algorm and o on. Becaue of hgh durably agan noe daa, good ably of generalzaon, hgh accuracy and low error rae, neural newor model poee grea advanage among daa mnng meod. w become a popular ool n daa mnng. 3. Daa Mnng Model of BP Neural Newor BP neural newor a nd of feedforward newor a now n mo common ue. Generally ha a mul-layer rucure a con of a lea ree layer ncludng one npu layer, one oupu layer and one or more hdden layer. There are full connecon beween neuron n adjacen layer and no connecon beween neuron n e ame layer. Baed on a e of ranng ample and a e of eng daa, BP neural newor ran neuron and complee e procedure of learnng. The applcaon of BP algorm uable for daa mnng envronmen n whch mpoble o olve problem ung ordnary meod. Therefore we need ue complex funcon of everal varable o complee non-lnear calculaon o accomplh e em-rucural and non-rucural deconmang upporng procedure. So n e procedure of daa mnng n e command cenre of polce offce, we chooe e BP neural newor model. The bac rucure of BP neural newor a follow: x y The learnng procedure of neural newor can be dvded no ree phae: () The fr one a forward propagaon phae n whch a pecfed npu paern ha been pa rough e newor from npu layer rough hdden layer o e oupu layer and become an oupu paern. () The econd one an error bac propagaon phae. In phae, BP algorm compare e real oupu and e expeced oupu o calculae e error value. Afer a, propagae e error value from oupu layer rough hdden layer o npu layer n e oppoe drecon. The connecon wegh wll be alered durng phae. Thee wo phae proceed repeaedly and alernaely o complee e memory ranng of newor unl end o convergence and e global error end o mnmum. 3.3 Learnng Algorm of Neural Newor In praccal applcaon of daa mnng, we ue e ree-layer BP neural newor model a nclude a ngle hdden layer and elec dfferenable Sgmod funcon a acvaon funcon. The funcon defned a formula (): f x) = + ( () e x The learnng algorm of BP neural newor decrbed a follow: W (0) () Seng e nal wegh value : Generally we generae random nonzero floang number n nal wegh value. [ 0,] a e () Choong ceran number of par of npu and oupu ample and calculae e oupu of newor. The npu ample are x = ( x, x, L, xn ). The oupu ample are = (,, L, m ), =,, L, L. L e number of npu ample. When e npu ample e ample e oupu of e neuron : x j y ) = f ( w ( ) x ) j j ( () Where e npu of neuron when e npu ample e j ample. j y x x n y m (3) Calculang e global error of newor. When e npu ample e ample e error of newor. The calculang formula of E E : Fg. The Srucure of BP Neural Newor

E ( ) = ( y ( )) = e ( ) Where repreen e neuron of oupu layer. (3) y () e oupu of newor when npu ample e ample and e wegh value ha been adjued me. Afer ranng newor me baed on all of e L group of ample, e global error of all of ee ample : G ( ) = E ( ) (4) Deermnng f e algorm end. (4) G () ε (5) When e condon of formula (5) afed e algorm end. ε e lm value of error a pecfed beforehand. ε > 0. (5) Calculang e error of bac propagaon and adjung e wegh. The graden decen algorm ha been ued o calculae e adjumen value of wegh. The calculang formula a follow: G( ) wj ( + ) = wj ( ) η wj ( ) E ( ) = wj ( ) η (6) w ( ) Where η learnng rae of newor and alo e ep of wegh adjumen. 3.4 The Problem of BP Newor and The Soluon Becaue we ue e graden decen algorm o calculae e value of wegh, BP neural newor ll encouner problem uch a local mnmum, low convergence peed and convergence nably n ranng procedure. We combne wo meod o olve ee problem. One oluon o mprove e BP newor algorm. By addng eep facor or acceleraon facor n acvaon funcon, e peed of convergence can be acceleraed. In addon, by compreng e wegh value when ey are oo large, e newor paraly can be avoded. The mproved acvaon funcon defned w formula (7): f x = a, b, λ ) + a ( x b + e ) λ j ( (7) a Where a devaon parameer, b a poon parameer and λ e eep facor. Anoer oluon a: Genec algorm a concurrence global earch algorm. Becaue of excellen performance n global opmzaon, o we can combne e genec algorm w BP newor o opmze e connecon wegh of BP newor. And fnally we can ue e BP algorm for accurae predcon or clafcaon. 4. UTILIZING GENETIC ALGORITHM TO OPTIMIZE BP NEURAL NETWORK 4. The Prncple of Genec Algorm Genec algorm a nd of earch and opmzaon model bul by mulang e lengy evoluon perod of heredy elecon and naural elmnaon of bologcal colony. I an algorm of global probably earch. I doen depend on graden daa and needn e dfferenably of e funcon a wll be olved and only need e funcon can be olved under e condon of conran. Genec algorm ha powerful ably of macro cope earch and uable for global opmzaon. So by ung genec algorm o opmze e wegh of BP neural newor we can elmnae e problem of BP newor and enhance e generalzaon performance of e newor. The ndvdual n genec pace are chromoome. The bac conuon facor are gene. The poon of gene n ndvdual called locu. A e of ndvdual conruc a populaon. The fne repreen e evaluaon of adapably of ndvdual o envronmen. The elemenary operaon of genec algorm con of ree operand: elecon, croover and muaon. Selec alo called copy or reproducon. By calculang e fne f of ndvdual, we elec hgh qualy ndvdual w hgh fne, copy em o e new populaon and elmnae e ndvdual w low fne o generae e new populaon. Generally ued raege of elecon nclude roulee wheel elecon, expecaon value elecon, pared compeon elecon and reanng hgh qualy ndvdual elecon. Croover pu ndvdual n populaon afer elecon no mach pool and randomly mae ndvdual n par o form paren generaon. Then accordng o croover probably and e pecfed meod of croover, exchange par of e gene of ndvdual a n par o form new par of chld generaon and fnally o generae new ndvdual. Generally ued meod of croover are one pon croover, mul pon croover and average croover. Accordng o pecfed muaon rae, muaon ubue gene w er oppoe gene n ome loc o generae new ndvdual. 4. The Calculang Sep of Genec Algorm The meod and ep of ulzng genec algorm o opmze e wegh of BP newor are decrbed a follow: () Fr, group of wegh are gven a random and agned o e of BP newor. By ranng e newor, group of new wegh ha been calculaed and adjued. They conue e orgnal oluon pace.

() Ung real number codng meod ee wegh are coded o decmal and ued a chromoome. group of chromoome compre a populaon. So e orgnal oluon pace ha been mapped o earch pace of genec algorm. The leng of gene rng afer codng. Where m e number of L = m h + h n neuron n npu layer, h e number of neuron n hdden layer and n e number of neuron n oupu layer. (3) Ung mnmum opmzaon meod e fne funcon can be deermned. The formula of fne funcon a follow: f = = G m = j= ( j y j ) (8) Where e oal number of ample, m e number of neuron n oupu layer, G e global error of all of number of ample and y j e oupu of newor. (4) The wegh are opmzed ung genec algorm. We calculae e fne and perform e elecon w meod of roulee wheel elecon. Afer a, we copy e ndvdual w hgh fne no nex generaon of e populaon. The nex ep croover. We croover e ndvdual afer elecon w probably P c. Becaue we ue real number codng meod o code wegh no decmal, e algorm of croover hould be alered. If e croover performed beween e ndvdual and e ndvdual, e operand a follow: ( +) ndvdual afer muaon, a random daum of unform drbuon n c [ u mn δ, umax + δ +. Afer once of ee operaon, a new populaon generaed. By repeang e procedure of elecon, croover and muaon, e wegh combnaon adjued cloe enough o e mo opmzed wegh combnaon. (5) Fnally, rough e BP newor e wegh can be adjued delcaely. Tll now, e whole procedure of opmzaon end. W repec o every nd of predcon and analy problem n e coure of daa mnng, we exrac proper e of ranng ample and eng daa, ran maure neural newor model w above-menoned meod and apply e model o e fuure cae analy and predcon. 5. A REAL INSTANCE OF APPLICATION Fnally, we gve a real applcaon of daa mnng n e command cener of polce offce a example. In example we analye people gender, age, educaon degree, hory of crme, experence, peronal feaure, ocal relaon and economcal ncome. And we fnd a o ome exen ee facor affec people ocal acon a may lead people o comm a crme. Ung ee facor a npu varable, a genec neural newor can be ulzed o predc e preen crme pobly of ee people. 5. Clean The Daa n Daabae In e fr ep, we fll up e mng daa, moo e noe daa n daabae and olve e problem of ame name for dfferen meanng and dfferen name for ame meanng. And en, we load relaed daa no daa warehoue. 5. Selec Tranng Sample of BP Newor ] + + + = c = ( c ) + ( c ) + c + + (9) Becaue n cae archve daabae e cae nformaon arranged n order of me, repreenave daa can be obaned by random amplng. So we elec ample by random amplng. To oban e ranng ample e of BP newor, we elec 5000 record from daa warehoue. In addon, we exrac oer 000 record a e eng ample e. Where croover, croover., + + +, + C 0,] Pm a par of ndvdual before a par of ndvdual afer a random daum of unform drbuon n [. W probably, we muae e ndvdual afer croover. If we muae e operand : = + c + ndvdual, e + (0) Where an ndvdual before muaon, an 5.3 rmalze Sample The mo mporan npu varable of BP newor nclude gender, age, educaon degree, crme hory, alary level and bad hab. The oupu of ample e au ( or ) of wheer people comm a crme a preen. The oupu of BP newor e probably of people preen au (Percenage) of crme. Table gve a l of fr 0 ample of e oal 5000 ranng ample.. Sex Age Educaon Degree M 5 Secondary chool M 3 Secondary chool Crme Hory Salary Level -- 300 Preen Bad Sau of Hab Crme

3 M 40 Prmary 4 F 30 Prmary 5 F 7 Secondary 6 M 8 Unvery 7 M Junor 50 Unvery 8 M 38 Pograduae 9 M 70 Prmary 0 F 35 Hgh 5000-- 0 500-- 3000 -- 500 5000-- 0 -- 300 Table Value of Inpu Varable By normalzng above npu and oupu varable, e range of value of ee varable ha been mapped o e range of [0, ]. The mappng relaonhp gven a follow: 5.3. Gender Male:.0; Female: 0.0 5.3. Age 0: 0.00; : 0.0; : 0.0; ; 00 and above:.0 4 0 0.45 0.5 0 0.75 0 0 5 0 0.7 0.5 0.5 6 0.8 0.65 0 0.5 0 0 7 0.50 0.5 0 0.375 0 0 8 0.38 0.75 0 0.75 0 0 9 0.70 0.5 0 0.5 0 0 0 0.35 0.375 0 0.5 0 0 Table rmalzed Value of Inpu Varable 5.4 Buld BP Neural Newor and Begn o Tran Becaue ncludng above 6 mporan varable e oal number of npu varable 0, we deermned a e number of neuron n npu layer 0 and e number of neuron n oupu layer. Accordng o our experence and conformng o e prncple of mplfyng e newor rucure, we e e number of neuron n hdden layer o 6. W above parameer we buld 0 BP newor a have ame rucure. Then we generae 0 e of mall random number a nal wegh of ee newor and ue e exraced 5000 ample a npu and oupu ample of ee newor. Afer a, we ulze BP algorm o ran e newor and ge 0 e of raned wegh. The ranng me are 0. Afer ranng we e e newor w our eng ample e. The generalzaon ably of our fr newor hown a follow: 5.3.3 Educaon Degree Illerae: 0.0; Graduae of Prmary : 0.5; Graduae of Secondary : 0.5; Graduae of Hgh : 0.375; Graduae of Junor Unvery: 0.5; Graduae of Unvery: 0.65; Pograduae: 0.75; Docor: 0.875; Po docor:.0 5.3.4 Crme Hory :.0; : 0.0 500 400 300 Y 00 G 00 G 5.3.5 Salary Level ne: 0.0; Below 300 Yuan: 0.5; 300 Yuan: 0.5; 500 Yuan: 0.375; 500 3000 Yuan: 0.5; 3000 5000 Yuan: 0.65; 5000 0 Yuan: 0.75; 0 5000 Yuan: 0.875; 5000 Yuan and above:.0 5.3.6 Bad Hab :.0; : 0.0 5.3.7 Preen Sau of Crme :.0; : 0.0 Table gve e value l of fr 0 ample of e oal 5000 ranng ample afer normalzaon.. Sex Age Educaon Degree Crme Hory Salary Level Bad Hab Preen Sau of Crme 0.5 0.5 0 0.5 0 0.3 0.5 0.5 3 0.40 0.5 0.5 000 4000 6000 0 Fg. The Generalzaon Ably of BP Newor Where me of ranng, Y e value of error, G global error of ranng ample e and G global error of eng ample e. 5.5 Ulze Genec Algorm o Opmze e Wegh Value We code e 0 e of raned wegh by real number codng meod and ue e wegh afer codng a chromoome. 0 group of chromoome con of a populaon. Then we opmze ee wegh ung genec algorm unl e wegh afer decodng are adjued cloe enough o e mo opmzed wegh combnaon. 5.6 Ue e Opmzed Wegh o Tran e BP Newor Agan Fnally, we ue one of e BP newor o adju e opmzed wegh delcaely. The ranng me for adjumen are

4000. A a reul, e generalzaon ably of e newor hown below: Y 50 40 30 G 0 G 0 000 000 3000 4000 Fg. 3 The Generalzaon Ably of BP Newor 5.7 Apply e Traned Newor o Predcon and Analy We ue e fnally adjued wegh a e runnng wegh of BP newor o predc e probably of crme a people may comm a preen. The probably e oupu of e BP newor and a floa pon number repreenng e occurrence probably of even. The predcon reul hghly accurae. In real wor of e command cenre of polce offce, predcon reul can be ued o gude e monorng and racng agan e former crmnal. A e ame me, can a e loc and confrmaon of upec n cae deecon. So hghly ueful for cae olvng and decon-mang. 6. CONCLUSION BP neural newor a ha been appled o daa mnng poee characerc of hgh ably of memory, hgh adapably, accurae nowledge dcovery, none rercon o e quany of daa and fa peed of calculaon. Ung genec algorm o opmze e BP newor can effecvely avod e problem of local mnmum. Therefore, e genec neural newor baed daa-mnng model ha many advanage over oer daa mnng model. In e real pracce of daa mnng n e command cenre of polce offce, e advanage have been fully emboded. L Yang, 004. A Daa Mnng Archecure Baed on ANN and Genec Algorm. Compuer Engneer, 30(6), pp. 55-56. Guo Zhmao, 003. An Exenble Syem for Daa Cleanng. Compuer Engneer, 9(3), pp. 95-96, 83 Qng Guofeng, 003. Acquremen of Knowledge on Daa Mnng. Compuer Engneer, 9(), pp. 0-. D.E.Goldberg, 99. Genec Algorm: A Bblography, IllGAL Techncal Repor, 90008. D.E.Goldberg, 989. Genec Algorm n Search, Opmzaon and Machne. Leanng, Addon-Weley. Srnva M., LAl M.Pana, 994. Genec Algorm: A Survey. IEEE Compuer, 7(6), pp. 7-6. Reference from Boo: Zhang Lmng, 993. The Model And Applcaon of Arfcal Neural Newor. Fudan Unvery Publher, Shangha. L Mngqang, 00. The Prncple and Applcaon of Genec Algorm. Scence Publher, Bejng. Davd Hard, 003. Prncple of Daa Mnng. Machne Indury Publher, Bejng. u Lna, 003. Neural Newor Conrol. Elecronc Indury Publher, Bejng. Beron Alex, Sm Sephen J. Daa Warehoung, Daa Mnng, &OLAP. McGraw-Hll Boo Co, 999. Reference from Journal: Aoyng Zhou, 005. A Genec-Algorm-Baed Neural Newor Approach for Shor-Term Traffc Flow Forecang. Advance n Neural Newor, 3498, pp. 965-969. Hecerlng Paul S, Gerber Ben S, 004. Ue of Genec Algorm for Neural Newor o Predc Communy- Acqured Pneumona. Arfcal Inellgence n Medcne, 30 (), pp. 7-75. u Zezhu, 004. A Daa Mnng Algorm Baed on e Rough Se Theory and BP Neural Newor. Compuer Engneer and Applcaon, 3, pp. 69-75. Wang Yu, 005. Predcve Model Baed on Improved BP Neural Newor and I Applcaon. Compuer Meauremen & Conrol, 3(), pp. 39-4.