Applicability of Backpropagation Neural Network for Recruitment Data Mining

Applicability of Backpropagation Neural Network for Recruitment Data Mining

Nitu Mathuriya Dr. Ashish Bansal SVIT, Indore SVITS, Indore

Abstract

One of the biggest issues in the industries and institutions is the selection of the right candidate for their organization. Recruitment is the big issue for any organization and this process consume the more time, efforts and investment of any organization. During each phase o f the recruitment process, candidates are filtered based on some performance criteria. The preliminary or initial process of recruitment in any organization is resume filtering or selection, any organization has many resumes and they have to select effective (or according to their use) resumes from lot of resumes .This preliminary process also need a educated staff (2 or 3 persons included) to collect or short out right resumes for the selection of the right candidate for the organization. The main objective of this work is initial process of the recruitment (resume screening or selection) is done using the neural network . In this research the domain k nowledge is extracted through knowledge acquisition technique. A Study has been made by applying back propagation neural network . Experiments were conducted by the data collected from the engineering institute to support their hiring power. The system supports managers in the process of assessment of specialists and managing the resources using formal methods. In this paper we describe experiments using artificial neural network s to automate the pre selection of the candidate to fill in the vacancies at their organization.

Key words:

Artificial neural network, knowledge acquisition, Industry, Organization, Pre selection

1 Intro ductio n

A frame work is proposed to imp rove the effectiveness of any organization in resume screening. In order to help large organizations (Industry and Institutes) manage resumes effectively, and improve the effic iency of recruit ment, the paper introduces a method that e xtract information fro m kinds of different data sets. The recruit ment process is one of the important functions of the HR depart ment in any organization and resume selection is the first step towards creating the good and effective staff for the organization. Resume selection process is very comple x and involves much of recourses and time o f the experts. This system is also useful to find out educated and effective staff for any institute and industry. The process involves lot of effort by the recruiting team and money spent for this process. Every organizat ion has its own recruit ment process and selection criteria. Rec ruit ment of the right and effective candidate is also important for IT companies to recruit the IT engineers, as well as the engineering institute. One of the mechanis ms used by the IT industries is to conduct test and group discussion, marks obtained in the semester exa ms and etc. On the other hand institutes has different criteria such as Marks obtained, experience, higher qualification, de mo, co mmun ication skill etc according to their or universities rules. So it is very difficult to imple ment a similar model fo r those. Resume filtering is a common process for any organization so its imple mentation is also beneficial to recru it ment department with fe wer e fforts. It has been observed that lot of resumes are collected only for 1 or 2 posts. Th e time taken to conduct interview of all the candidates consumes more t ime. And lot of effo rts is put to analyze the profile of the all applications to determine

the ones that suit the needs of the organization. Suppose 50applicants send their resumes for the 1 post means 1 a mong 50 applicants who apply get selected and the ratio of applicants interviewed after resume filtering is approximate 1:10,1:20,1:25 etc. Reducing this ratio will help the organization to s ave the effort.

Data min ing is the important approach to realize knowledge discovery. It is the process of extract ing knowledge or predicting previously unknown and useful trends from la rge quantities of data by using the knowledge of mult idisciplinary f ields such as statistics, mode identify, art ific ial intelligence, machine learning, database and so on. The artificia l neural network (ANN) is one of the most efficient techniques of data mining. It is the nonlinear auto-fit dynamic system made of many cells with simulating the construction of biology neural systems. ANN has the ability to mapping high nonlinear system, associable me mory and abstractly generalization. It can ma ke model fro m analyzing the mode in the data and discover the unknown knowledge. The present paper gives an engineering application of data mining based on neural networks. The back propagation (BP) neural network is used as the algorithm of data mining. In this paper, the back propagation (BP) neural network method is used as the technique of data mining to analy ze the effects of structural technologic parameters on efficiency in resume filtering. In this paper, results of the e xperiments conducted to cluster the data with the K means clustering and classification using back propagation algorithm have been analyzed.

The work reported in this paper attempts to perform the data mining task of Data Clustering. In the ANN paradigm, typically, supervised learning based BPN are used for data clustering tasks, because of their natural propensity to (a) find similarities amongst data items and (b) to group simila r data ite ms in pro ximity.

Traditional approaches to Informat ion Processing Vs. Neural Net works

1. TA : Simulate and forma lize hu man reasoning and logic process. TA treats the brain as a black box. A focus on how the elements are related to each other and how to give the mach ine the same capabilities.

   NN: Simu late the intelligence functions of the brain. NN focus on modelling the brain structure. NN attempts to create a system that functions like the brain because it has a structure simila r to the structure of the brain.

2. TA : The processing method of TA is inherently sequential.

   NN: The processing method of NN is inherently paralle l. Each neuron in a neural network system functions in paralle l with others.

3. TA : Learning takes place outside of the system. The knowledge is obtained outside the system and then coded into the system.

   NN: Learn ing is an integral part of the system and its design. Knowledge is stored as the strength of the connections among the neurons and it is the job of NN to learn these weights from a data set presented to it.

4. TA: Is deductive in nature. The use of the system involves a deductive reasoning process, applying the generalized knowledge to a given case.

   NN: Is inductive in nature. It constructs an internal knowledge base from the data presented to it. It generalizes fro m the data, such that when it is presented a new set of data, it can make a decision based on the generalized internal knowledge

5. TA: It represents knowledge in an explicit form. Rules and relat ionships can be inspected and altered.

NN: The knowledge is stored in the form of interconnections strengths among neurons. Nowhere in the system, can one pick up a piece of computer code or a numerica l value as a discernible piece of knowledge.

1. Data mining Techni ques

   The machine learn ing and the statistics are two main technical approaches of data min ing. The machine learning, as a broad subfield of artificia l intelligence, is concerned with the development of algorith ms and techniques that allow computers to learn ability to achieve the tasks of identifying, inducing, classification, predicat ion etc. Artific ial Neura l

   Network (ANN) and K means are the most widely applied methods in those fields. ANN is an informat ion processing paradigm that is inspired by the way biological nervous systems, such as the brain, process informat ion. It is composed of a large number of highly interconnected processing elements (neurons) working in unison to solve specific proble ms. The most typical neural networks are the BP neural network, the Hopfie ld neural networks and the adaptive neural networks. As the other technical support of data mining, the statistics offers the most fundamental theory of data mining techniques based on the precise mathe matica l approach.

2. Neur al Ne twork techni ques in data mining

   A Neural Network is a powerful data-modelling tool that is able to capture and represent comple x input/output relationships. The motivation for the development of neural network technology stemmed fro m the desire to develop an artific ia l system that could perform "intelligent" tasks simila r to those performed by the human brain. Neura l networks resemble the hu man brain in the fo llo wing two ways:

   A neural network acquires knowledge through learning. A neural network's knowledge is stored within inter- neuron connection strengths known as synaptic weights. The Operation of a Neura l Network is controlled by three properties: The pattern of its interconnections, architecture. Method of determin ing and updating the weights on the interconnections, training. The function that determines the output of each individual neuron, activation or transfer function. In data warehouses, neural networks are just one of the tools used in data min ing. ANNs are used to find patterns in the data and to infer rules fro m the m. Neura l networks are useful in providing infor mat ion on associations, classifications, clusters, and forecasting. The back propagation algorithm performs learn ing on a feed-forward Bac k propagation neural network. A neural network has to be configured such that the application of a set of inputs produces (either 'direct' or via a rela xation process) the desired set of outputs. Various methods to set the strengths of the connections e xist. One way is to set the weights explic itly, using a priori knowledge. Another way is to 'train' the neural network by feeding it teaching patterns and letting it

   change its weights according to some learn ing rule. Supervised learning or Associative learn ing which incorporates an external teacher, so that each output unit is told what its desired response to input signals ought to be. During the learning process global informat ion may be required. Paradig ms of supervised learning include error-correction learning (back propagation algorithm), reinfo rce ment lea rning and stochastic learning. The input-output pairs can be provided by an external teacher, or by the system which contains the neural network (self-supervised).

3. Back propag ation Ne ural Network Techni que

   Back propagation is a form of supervised learning for mu lti-layer nets, also known as the generalized delta rule. Error data at the output layer is "back propagated" to earlie r ones, allowing inco ming we ights to these layers to be updated. It is most often used as training algorith m in current neural network applications. The back propagation algorithm was developed by Paul Werbos in 1974 and rediscovered independently by Ru me lhart and Parke r. Since its rediscovery, the back propagation algorithm has been widely used as a learning algorith m in feed forward multilayer neural networks. What makes this algorithm diffe rent than the others is the process by which the weights are calculated during the learning network. In general, the difficulty with mult ilayer Perceptrons is calculating the weights of the hidden layers in an efficient way that result in the least (or zero) output error; the more hidden layers there are, the more difficu lt it becomes. To update the weights, one must calculate an error. At the output layer this error is easily measured; this is the diffe rence between the actual and desired (target) outputs. At the hidden layers, however, there is no direct observation of the error; hence, some other technique must be used. To calculate an error at the hidden layers that will cause minimizat ion of the output error, as this is the ultimate goal.

   The back propagation algorithm is an involved mathe matica l tool; however, e xecution of the train ing equations is based on iterative processes, and thus is easily imp le mentable on a co mputer.

4. Algorithm

1. Present a training sa mple to the neural network.

2. Co mpare the network's output to the desired output fro m that sample . Calcu late the error in each output neuron.

3. for each neuron, calculate what the output should have been, and a scaling factor, how much lowe r or higher the output must be adjusted to match the desired output. This is the local error.

4. Adjust the weights of each neuron to lower the local error.

Actual Algorithm:

1. Init ia lize the weights in the network (often randomly )

2. Repeat

   * for each e xa mple e in the tra ining set do

   1. O = neural-net-output (network, e) ; Forward pass

   2. T = teacher output for e

   3. Ca lculate error (T – O) at the output units

   4. Co mpute delta_wi for all we ights from hidden layer to output layer;

      Backward pass

   5. Co mpute delta_wi for all we ights from input layer to hidden layer;

      Backward pass continued

   6. Update the weights in the network

* end

1. until all e xa mp les classified correctly or stopping criterion satisfied

2. Return (network)

1. Proposed Model

   Neural networks, with their re ma rkab le ability to derive mean ing fro m co mplicated or imp recise data, can be used to detect trends that are too comple x to be noticed by either humans or computer techniques. The application of data mining based on neural networks can be generally divided into stages: Knowledge acquisition, data preparation, and modelling and knowledge discovery, as showed in Figure 1.

   The design of the system requires the complete understanding of the problem do ma in. The data sets and the input attributes are determined through knowledge of an engineering college. Data preparat ion is the definit ion and e xpression of the mined data, which can ma ke the mined data suitable for the algorith m. Data preparation is the most important step of data min ing. Data selection means selecting data useful to data mining. It includes selections in two dimensions of line and row. First, the line dimension selection, namely fie ld dimension, determines the neuron nodes of input and output layers of the neural network. Second, the row d imension selection, na mely record dimension, determines the samples used to train and check the neural network. Be fore the selected data are used in the mining, they should be converted to a form acceptable, which is data expression. First, symbolic data type should be converted to a numerica l one because neural network mode ls can only deal with numerical data type. On the other hand, neural network models only accept the data between 0 and 1 or between -1 and 1 as input data. Therefore, the data should be mapped to this region according proportional transformation or others. The task and purpose should be confirmed at the beginning of data-mining stage. What algorithm would be used in mining is determined according to the characteristics of the task type. The back propagation algorithm is a widely used method in neural networks. It is e xtensively used in pattern recognition, character e xtraction, etc. After the determination of neural network models, network training can be conducted by using the selected data. This is a repeated process, in which the architecture of the neural network is determined. The aim is to determine a set of weights and threshold values that satisfy the precision require ment.

   Knowledge Acquisition from domain experts

   Data Preparation

   www.ijert.org 4

   informat ion and Experience. Pe rsonal informat ion included Candidate na me, address, mob ile number and Ema il id.

   Table 1: Personal information of the candidates (data set 1)

   S no.	Na me	Address	Mobile no	Email id
   1.	xyz	Indore	9926645442	xyz@gmail.com
   2.	abc	Dewas	6548125452	abc@gmail.com
   3.	prq	Indore	5468544252	prq@yahoo.com
   4.	mnn	Indore	9926542584	mnn@rediffmail.com
   5.	s tv	Ujjain	9300258244	stv@gmail.com

   Table 2: Educational qualification of the candidates (Data Set 1)

   BE Year	Branc h	BE %	ME year	ME %	PHD
   2005	CS	70	2010	72	Yes
   2011	CS	60	No	No	No
   2006	IT	72	2008	76	No
   2005	EC	68	2007	72	No
   2012	ME	74	NO	No	No

   Figure 1 Data mining base d on neur al ne tworks

   Finally it is necessary that the discovered knowledge should be visualized, which can make the rules understandable easily and beneficia l to applications. The constructed models were reviewed and evaluated before it is used for selection. The models were evaluated using accuracy as the criteria to assess the performance of the method.

2. Experime ntal Conside ration

   The system proposed in this paper has been imple mented and evaluated with e xtensive e xperimentations on the collected data sets from engineering college for rec ruiting teaching staff in the college. Accuracy of the classification is used as a metric for performance. This section presents with the details of the data sets.

   The data set of table 1 contains the all informat ion of the candidate e. g. personal information, Educational

   Table 3: Experience and Extra qualification of the candidates with selection output (Data Set 1)

   Industry Experience	Teaching Experience	Extra Qualification (Gate card)	Desired Output
   2	2	Yes	1
   No	1	No	0
   3	1	No	1
   No	4	Yes	1

   No

   No

   No

   0

   The data set contains these input tables data in a one table with the desired output. Firstly apply the data selection and data expression on these data set1.After the data pre-processing apply the Back propagation technique on these data set. The BP neura l network model consists of three layers. The topology of the neural network model is shown in Figure 2.

   Figure 2 Architecture of Neur al Network

   After the sample tra ining, a set of weights and threshold values which satisfy the precision require ment are achieved. All computer programs are completed in the environment of MATLAB. The results of data min ing are shown in table 4.

   Table 4: Result1 (Output table)

   S no .	Name	Address	Mobile no	Email id
   1.	xyz	Indore	9926645442	xyz@gmail.com
   3.	prq	Indore	5468544252	prq@yahoo.com
   4.	mnn	Indore	9926542584	mnn@rediffmail.c om

   This output table shows the personal informat ion of the selected candidates from the data set1.Table 4 g ive the informat ion of those candidates after applying the neural network technique whose resume informat ion deserve for the selection. This filtering done by applying the BPNN technique on training data set and finds the result.

3. Evaluation Methodology

   The metric used to evaluate the clustering and classification algorith m is the accuracy. Accuracy is determined as the ratio of records correctly classified

   during testing to the total number of records tested. BPNN techniques were applied with MATLAB and the accuracy of the network is depicted in table 5.It is observed that neural network techniques have better accuracy and suitable for this proble m do ma in.

   Table 5: Result of BPNN technique trained with data set1 and tested with dataset2

   Training data set	Test Data set	Test data set Record	Accuracy
   Dataset1	Dataset1	450 approx	92
   Dataset1	Dataset2	300 approx	88
   Dataset2	Dataset 2	300 approx	83
   Dataset2	Dataset1	450 approx	78

   Dataset 1 and Dataset 2 both contain different record sets. It was observed that the accuracy of BPNN better than other traditional a lgorith ms.

4. Conclusion and Future work

In this paper we present a neural network based approach to mining c lassification ru les fro m g iven databases. The approach consists of diffe rent phases:

1. Constructing and training a network to correctly classify tuples in the given training data set to required accuracy,

2. Extract ing knowledge through the network using BPNN

A set of experiments was conducted to test the proposed approach is using a well defined set of data mining proble ms. The results indicate that, using the proposed approach, high quality or useful data can be discovered from the given data sets. In future we can apply this technique to select deserving candidates for an organization. This technique is also applicable on feedback system in any organization. Artificia l Neura l Networks offer qualitative methods for business and economic systems that traditional quantitative tools in statistics and econometrics cannot quantify due to the comple xity in translating the systems into precise mathe matica l functions. Hence, the use of neural networks in data mining is a pro mising fie ld of research especially given the ready availability of large mass of data sets and the reported ability of neural networks to

detect and assimilate relat ionships between a large numbers of variab les.

References

1. Jiawei Han, M icheline Kamber. 2006. Data M ining Concepts and Techniques, Second Edition M organ Kaufmnn Publishers, San Francisco.

2. Haykin, S., Neural Networks, Prentice Hall International Inc.

3. L. Wang and T. Z. Sui, Data M ining Technology Based on Neural Network in the Engineering, School of M echanical Engineering and Automation ortheastern University IEEE 2007

4. N. Sivaram Research Scholar and V, Department of CSE ,Clustering and Classification Algorithms for Recruitment Data M ining, International Journal of Computer Applications (0975 8887) Volume 4

   No.5, July 2010

5. Artificial Neural Network, Wikipedia Encyclopedia, Wikipedia Foundation, Inc., 2006.

6. Arun K Pujari ,Data mining techniques universities press India(Private limited) Eight Impression 2005