Enhancing Graduate Academic Performance Prediction and Classification: An Analysis Using the Enhanced Correlated Feature Set Model

In the face of escalating pressures from globalization and domestic economies, educational institutions are being compelled to implement innovative changes aimed at enhancing the quality of their academic programs and increasing the graduation rates of students with specialized degrees [1]. Students, being the primary investors in these institutions, substantially contribute to a country's economic and social development through the production of creative graduates. Pedagogical institutions hold a vested interest in maintaining operability and managing large student databases for efficient administration. With the proliferation of web technology, its adoption has become a standard practice in many educational institutions, facilitating the sharing of vital data about students, faculty, and their engagement with educational systems [2]. Supporting the development and expansion of society, higher education provides a community rich in resources, including directories of universities, faculty, staff, and listings of available training and development opportunities [3].

Academic success is a shared interest among educators, institutions, and businesses, and as such, it is crucial that students strive to excel acadically to meet these stakeholders' expectations [4]. The reputation of a university grows in tandem with the quality of its graduates, making academic success a pivotal factor [5]. While the economic ramifications and extracurricular activities of a student may influence their academic success, most research posits that the culmination of academic success is marked by graduation [6].

The transformation of large data volumes into useful information for individuals, organizations, and governments necessitated the integration of previous methodologies. Data scientists rely heavily on insights derived from accounting records [7], empirical studies, and predictive models. The focus has now shifted towards predictive and correlative analytics, thanks to the development of efficient data storage, retrieval, and management systems [8]. Information science can be leveraged by educators to evaluate student progress in comparison to their peers, enhancing the systems of tutoring and monitoring employed by administrators and leaders [9].

Students embarking on their college journey at an introductory level often encounter multiple obstacles that may impede their academic progress. Not all students who commence higher education at a freshman level successfully graduate with the required coursework. Many university professors struggle to provide adequate support to their students. This work focuses on advanced organizational strategies for providing relevant conceptual support and meaningful verbal material, defining feedforward teaching methodologies to enhance learning. However, there is a paucity of research on feedforward strategies to improve students' learning outcomes. The aim of this study is to lay the groundwork for improved academic support and student autonomy. The primary contributions of this paper are the identification of critical feedforward properties and the proposal of effective feedforward strategies, with the objective of systematically adopting these strategies to aid student populations at the introductory level.

The prediction of academic performance is a central problem in the field of education data mining, as it is an essential precursor to achieving personalized learning [10]. Multiple studies have demonstrated that the following factors significantly influence academic performance:

• Personality traits of the students, such as neuroticism, extroversion, and agreeableness.
• Personal situations.
• Motivating factors in daily life.
• Learning-promoting behaviors.

Institutional managers can glean insights into a student's likelihood of graduation by analyzing academic performance data [11]. Traditionally, this analysis is conducted by professors, who, through class activities and midterm assessments, identify students at risk and implement timely corrective measures [12]. However, in contemporary higher education, identifying such students has become increasingly challenging due to escalating class sizes and diminished opportunities for individual student-faculty interaction. This is exacerbated by the increased availability of online educational resources and the growing number of students who are concurrently employed full-time [13].

Data mining and deep learning analyses provide more precise forecasting of student outcomes. With the advent of widespread computerization, many educational institutions amass vast amounts of data, much of which remains unutilized [14]. Extracting value from such data necessitates robust tools [15], and deep learning algorithms fit this role.

Feature selection is a crucial component of developing a student performance prediction system [16]. It can enhance the accuracy of academic performance prediction and identify the key factors influencing student performance. A novel score and ranking system is employed to develop a set of candidate qualities, followed by using a heuristic approach for the final result [17]. Predicting a student's academic performance can assist universities in identifying individuals who require financial aid, enhancing the quality of incoming students, facilitating students’ future planning, and helping struggling students identify solutions. The model used for performance prediction is driven by the attributes selected from the dataset [18]. Teachers can choose the most useful features using a feature selection technique, improving the outcomes of predictive analyses. Feature selection algorithms are particularly beneficial in educational institutions as they excel at extracting key features and avoiding redundancy without the risk of data loss [19]. Feature selection methods are employed to increase prediction accuracy while minimizing computational complexity, potentially enhancing the accuracy of models used to predict student grades [20].

The fields of Deep Learning and data mining in education have received substantial attention in recent years. This research proposes a Deep Neural Network (DNN) model to assist students in choosing appropriate class labels, aiding schools in identifying and supporting students at risk of failing [21]. Machine learning programs, utilizing students' historical data and term exam results, have proven effective in predicting future course performance. With these techniques, educators can swiftly identify students at risk of failing and intervene to enhance their performance. Moreover, it can identify top-performing students for scholarship opportunities.

Student's academic standing and progress are reflected in grades, rankings, and passing standards [22]. Predicting student performance benefits academia monitoring and early warning systems, enabling students to make timely adjustments to their study habits and progress reports to avert setbacks [23]. Teachers can rapidly adapt their lesson plans to suit the needs of their classes and improve passing rates. Increased attention from counselors to students' needs can lead to improved academic outcomes. The future success of college and university graduates may hinge on differentiated teaching [24]. A strong correlation between third-year and final year performance has been observed using the proposed Graduate Interlinked Precedent Academic based Performance Analysis using Enhanced Correlated Feature Set Model.

Recently, a new educational philosophy known as outcome-based education (OBE) has gained widespread support and popularity. When it comes to education, this new paradigm puts the emphasis on what students actually learn rather than on what teachers originally set out to do. The phrase student outcome is commonly used to describe the expected levels of student mastery of content and transferable skills and values by the end of a course or program [25]. Course outcomes and program outcomes are two possible levels at which the outcomes are defined and evaluated. Curriculum mapping is an essential activity because it ensures that courses and programs are aligned in a way that allows students to successfully complete their intended learning outcomes [26].

Digital tools were developed to more effectively record the educational evaluation activities and to promote the accomplishment of the OBE goals. The value of these tools could be greatly enhanced with the addition of intelligent models that can anticipate whether or not students will accomplish learning outcomes during the course of an education. The value of tracking student progress in higher education is difficult to overstate. These include, but are not limited to, defining the expected outcomes of the program, outlining expectations for students and teachers, and evaluating the efficacy of individual classes and entire programs. Several quality evaluation instruments and quality management frameworks were presented to facilitate the implementation of the outcome-based education paradigm and the accreditation of programs.

Predicting students' performance provides other benefits, such as the chance to apply remedial interventions in the teaching and learning processes. Nonetheless, there is a lack of publications that investigate intelligent prediction of educational results in depth. Furthermore, there is a lack of information about the factors and traits that affect academic achievement. Both academic factors, like the quality of education and the amount of time children spend online, and extracurricular factors, such parental participation and the level of internal drive demonstrated by students, are important, as shown by the existing studies. The dataset used in this study is comprised of information collected from universities about their students' academic performance.

Labeled data is data for which the desired outcome has already been determined. When performing feature extraction, users are essentially paring down a set of attributes. As opposed to feature selection, which merely ranks the current attributes in order of their predictive significance, feature extraction actively modifies the attributes. The features that result from the transformation are linear combinations of the original features. The models will be evaluated using the remaining labelled data we have. Before further processing, some deep learning methods were applied to categorize the raw data.

Students' attitudes, knowledge, and abilities must be constructed in a true and comprehensive method based on student activities; hence, evaluation cannot be done solely in the cognitive domain. Competence, attitudes, knowledge, and skills are all assessed in a balanced manner so that each student's progress in relation to predetermined goals can be determined. As a result, the old system of evaluating students' performance gave way to a more authentic evaluation system (actual), with assessment of students' psychomotor abilities becoming a key component of this latter. Performance evaluation is one type of genuine assessment that might be used to offer comprehensive data on students' progress toward and actual attainment of their Physics learning outcomes. Students' abilities in many domains can be evaluated through performance appraisal by proving their proficiency in these domains. One alternate approach to gauging students' knowledge and abilities in the actual world is through the use of performance-based assessments. Since these assessments measure the development of learning not only via results but also processes and varied ways, performance appraisal is an element of a real evaluation that is thought to be able to better quantify the total results of student learning. When it comes to the psychomotor domains of physics learning, evaluation of performance is one of the most highly suggested assessment method. One other method of evaluation is the performance review. As a result, this method of evaluating students' progress can contribute to their learning.

A key aspect of any model-building effort is an evaluation of the working model. It aids in determining which model most accurately depicts available data and how effectively the selected model will function in the long run. In data science, it is not acceptable to evaluate a model's efficacy using the same data that was used to train it, as doing so can easily lead to overoptimistic and over fitted models. Every classification model's performance is estimated by taking the mean. Using graphs to display information that has been previously categorized. Accuracy is the proportion of times test data is predicted correctly. A strong correlation between third-year and final year performance has been found using the proposed Graduate Interlinked Precedent Academic based Performance Analysis using Enhanced Correlated Feature Set Model (GIPA-PA-ECFSM) model.

The students set {S₁, S₂,………….., S_M} is considered and the academic parameters { P₁, P₂……., P_N} are considred for the academic performance analysis. The average academic parameters are considered as Eq. (1):

$P_{\text {analysis }}=\frac{\sum_{s=1}^M \operatorname{maxgrade}\left(P_s\right)+\operatorname{maxgrade}\left(P_{s+1}\right)}{\sum_{s=1}^M \operatorname{maxgrade}\left(P_{s+N}\right)}$        (1)

The features from the dataset are considered that is used for training the model. The features extracted will be used for academic performance analysis. The feature extraction is performed as Eq. (2):

$F \operatorname{set}(S G [M])=\sum_{s=1} \frac{\operatorname{avg}\left(\max \left(P_s, P_{s+1}\right)+\operatorname{getattr}\left(P_{,} P+S\right)\right)}{\operatorname{size}(\lambda)}+\frac{\min \left(P_s, P_{s+1}\right)}{\operatorname{size}(\lambda)}$        (2)

Here λ is the records in the dataset considered. The features extracted are used to verify the correlation factor and the maximum correlated values are considered for analyzing the performance levels. The enhanced correlated feature set model is generated as Eq. (3):

$\begin{aligned} & \text { CFset }(S(i))= {\left[\sum_{s=1}^M+\frac{\left(\max (\operatorname{Semester}(P(s+1))-\max (\operatorname{semester}(s)))^N\right.}{\min (\operatorname{semester}(s)) / \min (\operatorname{semster}(s-1))}\right]}\end{aligned}$           (3)

A deep learning model is applied with deep layered structure representation of performance analysis parameters as layers that can improve accuracy rate and level ordering. Estimating the appearance of a permutation layer allows us to join two academic parameters sections that are separated by it. Different convolution kernels can be detected in extracted feature mapping sets like map P and map Q if they are obtained by summing their convolutions. Common practice has it that the features are used as input, and the academic performance levels in each emester is then added up ana analyzed for final prediction. The hidden layer processing is performed as Eq. (4):

$\begin{aligned} & \text { Hidd_Layer }(\mathrm{i}, \mathrm{o}, \mathrm{P})=\sum_{p=1}^M \max (F \operatorname{set}(p)) +\frac{\max (C F \operatorname{set}(p)}{\min (C F \operatorname{set}(p)} * \operatorname{len}(C F \operatorname{set})+\frac{\min (P(i, i+1))+\lim _{s \rightarrow p}\left(\max (p)+\frac{\min (p)}{M}\right)^2}{\operatorname{sizeof}(p)}\end{aligned}$            (4)

The performance levels of the undergraduates are analyzed based on the features considered. These features are the mostly useful features in identification of student academic performance. The feature subset interlinking is performed as Eq. (5):

$\begin{aligned} & \text { Flink }(\operatorname{CFset}[M])= & \frac{\sum_{s \in \text { CFset }} \operatorname{mean}\left(P_s, P_{s+1}\right)+\text { Hidd_Layer }(s)-\sum_{s=1} \min \left(P_{s-1}, P_s\right)}{\text { size }(\text { CFset })} \left\{\begin{array}{lc}\text { if } \quad \text { mean }\left(P_s, P_{s+1}\right)> & Th ~\text {Im } proved \\ \text { Otherwise } & \text { Poor }\end{array}\right\}\end{aligned}$            (5)

The proposed model performs the student performance analysis based on the interlinking features for accurate academic performance levels. The academic performance analysis is performance as Eq. (6):

$\begin{aligned} & \text { Perf_Pred }[M]=\frac{\sum_{s=1}^M \max (F \operatorname{link}(i))+\operatorname{diff}(\max (P(s+1), P(s))}{\max (F \operatorname{link}(s))} +\frac{\sum_{s=1}^M \operatorname{getatt}(P(s-1), P(s))}{\max (F \operatorname{set}(s))}\end{aligned}$            (6)