Evaluation Of Portugese Students' Progress In The Schools Of Secondary Education Using Business Intelligence And Data Mining Techniques

Abstract

Despite of the education levels in the Portuguese population has success in improving in the last decades, according to statistics keep Portugal at the tail of the European continent that's due to the very high student failures, especially the classes of lack of success are the Portuguese language and mathematics and they are extremely dangerous on the student levels.

From that point the Data mining and business intelligence specialists tried to get their knowledge in that matter and get more valuable insights from data about students. The current work aims to approach students' progress in the schools of secondary education using business intelligence and data mining techniques. And we collected the existed data in the data about everything related to the students about their progress in school and their family's status. and that data was collected by school questionnaires and students the both main classes (Math and Portuguese language) were modeled by the way of binary/five-level classification and regression techniques. And for data mining models like (Regression, SVM, neural networks, and Decision tress). Despite of student achievement is highly affected by past evaluations, a data exploratory analysis has revealed that there are also other some relevant features like (number of absences, parent’s job and education, alcohol consumption). As a direct result of that search, more efficient student prediction models can be developed, to improve the quality of education and enhance the resource management of the schoolI.

Introduction

The education is a fundamental factor to get a long haul advance in every one of the fields. At the season of the last several years, the Portuguese instructive level has moved forward. Nonetheless, the insights keep the Portugal at Europe's last part because of its high understudy disappointment and dropping out rates. For instance, in 2006 the early school leaving rate in Portugal was 40% for 18 to 24 years of age, while the European Union normal esteem was only 15%. Specifically, disappointment in the center classes of Mathematics and Portuguese (the local dialect) is to a great degree genuine, since they give major learning to the achievement in the rest of the school subjects (e. g. material science or history). Then again, the enthusiasm for Business Intelligence (BI)/Data Mining (DM), emerged because of the advances of Information Technology, prompting an exponential development of business and authoritative databases. This information holds significant data, for example, patterns and examples, which can be utilized to enhance basic leadership and enhance achievement. However, human specialists are restricted and may disregard vital points of interest. Henceforth, the option is to utilize robotized devices to break down the crude information and concentrate intriguing abnormal state data for the leader.

Literature review and proposed plan

In actuality, a few investigations have tended to comparable themes. Maet al. (2000) connected a DM approach situated in Association Rules with the end goal to choose feeble university understudies of Singapore for medicinal classes. The information factors included statistic properties (e. g. sex, locale) and school execution over the previous years and the proposed arrangement outflanked the customary designation strategy. In 2003 online understudy grades from the Michigan State University were displayed utilizing three grouping approaches (i. e. paired: pass/come up short; 3-level: low, center, high; and 9-level: from 1 - least grade to 9 - most noteworthy score). The database included 227 examples with online highlights (e. g. number of remedied answers or strives for homework) and the best outcomes were acquired by a classifier outfit (e. g. Choice Tree and Neural Network) with exactness rates of 94% (parallel), 72% (3-classes) and 62% (9-classes).

Kotsiantis et al. (2004) connected a few DM calculations to anticipate the execution of software engineering understudies from a college separate learning program. For every understudy, a few statistic (e. g. sex, age, conjugal status) and execution qualities (e. g. markin a given task) were utilized as contributions of a paired pass/fizzle classifier. The best arrangement was gotten by a naive Bayes strategy with a precision of 74%. Likewise, it was discovered that past school grades have a significantly higher effect than statistic factors.

All the more as of late, Pardos et al. (2006) gathered information from an internet mentoring framework with respect to USA eighth grade Math tests. The creators embraced a relapse approach, where the point was to anticipate the math test score dependent on individual abilities. The creators utilized Bayesian Networks and the best outcome was a prescient mistake of 15%. MethodologyIn this work, we will break down ongoing genuine information from two Portuguese optional schools. Two unique sources were utilized: stamp reports and surveys. Since the previous contained rare data (i. e. just the evaluations and number of unlucky deficiencies were accessible), it was supplemented with the last mentioned, which permitted the accumulation of a few statistic, social and school related characteristics (e. g. understudy's age, liquor utilization, mother's instruction). The point is to foresee understudy accomplishment and if conceivable to distinguish the key factors that influence instructive achievement/disappointment. The two center classes (i. e. Science and Portuguese) will be demonstrated under three DM objectives:

1. binary classification (pass/fail);
2. classification with five levels (from I very good orexcellent to V - insufficient);
3. regression, with a numeric output that ranges between zero (0%) and twenty (100%).

For every one of these methodologies, three information setups (e. g. with and without the school time frame evaluations) and four DM calculations (e. g. Decision Trees, Random Forest) will be tried. In addition, a logical examination will be performed over the best models, with the end goal to distinguish the most important highlights.

Data Mining Models

Regression and classification are two important Data mining goals. Both required for a supervised learning, where a modelis suitable to a dataset made up of k ∈ {1, . . . , N} examples, each mapping an input vector (xk 1, . . . , xk I) to a given target yk. The main change is set in terms of the result representation, (i. e. discrete for classification and continuous for regression). In classification, technique are often evaluated using the Percentage of Correct Classifications (PCC), while in regression the Root. Intuitively, the work of directors, writers and Mean Squared (RMSE) is a popular metric. A high PCC (i. e. near 100%) suggests agood classifier, while a regressor should present a lowglobal error (i. e. RMSE close to zero). These metricscan be computed using the equations:Φ(i) = 1 0, , if else yi = ybiPCC = PN i=1 Φ(i)/N × 100 (%)RMSE = qPN i=1 (yi - ybi)2/N(1) where ybi denotes the predicted value for the i-th example. In this work, the Mathematics and Portuguese grades(i. e. G3 of Table 1) will be modeled using three supervised approaches:

1. Binary classification – pass if G3≥10, else fail.
2. 5-Level classification – based on the Erasmus1 grade conversion system.
3. Regression – the G3 value (numeric output between 0 and 20).

Many DM algorithms, each one with its own purposesand capabilities, have been proposed for classificationand regression tasks.

In Portugal, the optional instruction comprises of 3 years of tutoring, going before 9 years of essential training and pursued by advanced education. The greater part of the understudies join people in general and free instruction framework. There are a few courses (e. g. Sciences and Technologies, Visual Arts) that offer center subjects, for example, the Portuguese Language and Mathematics. Like a few different nations (e. g. France or Venezuela), a 20-point reviewing scale is utilized, where 0 is the most reduced review and 20 is the ideal score. Amid the school year, understudies are assessed in three periods and the last assessment relates to the last grade. This examination will consider information gathered amid the 2005-2006 school year from two state funded schools, from the Alentejo area of Portugal. Despite the fact that there has been a pattern for an expansion of Information Technology venture from the Government, most of the Portuguese state funded school data frameworks are extremely poor, depending for the most part on paper sheets (which was the present case). Henceforth, the database was worked from two sources: school reports, in view of paper sheets and including few qualities (i. e. the three period grades and number of school absences); and surveys, used to supplement the past data. We composed the last with shut inquiries (i. e. with predefined alternatives) identified with a few statistic (e. g. mother’s education, family income), social/enthusiastic (e. g. alcohol consumption) and school related (e. g. number of past class failures) factors that were relied upon to influence understudy execution. The survey was checked on by school experts and tried on a little arrangement of 15 understudies with the end goal to get an input. The last form contained 37 inquiries in a solitary A4 sheet and it was replied in class by 788 understudies. Last mentioned, 111 answers were disposed of because of absence of recognizable proof subtle elements (important for converging with the school reports). At long last, the information was coordinated into two datasets identified with Mathematics (with 395 models) and the Portuguese dialect (649 records) classes.

Amid the preprocessing stage, a few highlights were disposed of because of the absence of discriminative esteem. For example, couple of respondents replied about their family wage (most likely because of security issues), while very nearly 100% of the understudies live with their folks and have a PC at home.

Data Challenges

We have transferred the data from string to one hot encoder for labeling the data because the algorithm doesn't deal with strings and by that we could handle data and get the accuracy of the algorithm very perfect. The main core class (i. e. Mathematics) will be modeled under three DM goals:

1. binary classification (pass/fail);
2. classification with five levels (from I very good orexcellent to V - insufficient);
3. regression, with a numeric output that ranges between zero (0%) and twenty (100%).

Conclusion

Education is a crucial element in our society. Business intelligence (bi)/data mining (dm) techniques, which allow a high-level extraction of knowledge from raw data, offer interesting possibilities for the education domain. In particular, several studies have used bi/dm methods to improve the quality of education and enhance school resource management. In this paper, we have addressed the prediction of secondary student grades of on main class (mathematics) by using past school grades (first and second periods), demographic, social and other school related data. Three different dm goals (i. e. Parallel/5-level classification and regression) this study was based on an off-line learning, since the dm techniques were applied after the data was collected. Notwithstanding, there is a potential for an automatic on-line learning environment, by using a student prediction engine as part of a school management support system. This will make collection of additional features (e. g. Evaluations from previous school years) and also to obtain a valuable feedback from the school professionals. Moreover, we intent to enlarge the experiments to more schools and school years, in order to enrich the student databases. Programmed feature selection methods (e. g. Separating or wrapper) will also be explored, since only a small portion of the input variables considered seem to be relevant. In particular, this is expected to benefit the nonlinear function methods (e. g. Nn and svm), which are more sensitive to irrelevant inputs. More research is also needed (e. g. Sociological studies) in order to understand why and how some variables (e. g. Motivation to choose school, parent's job or alcohol consumption) affect student performance.