Factors Affecting the Academic Performance of Students with Hearing Impairment

According to World Health Organization’s (WHO) World Report on Disability 2011, people with disabilities make up around 15% of the global population. As per WHO, hearing loss will affect over 900 million people by 2050, up from 450 million today, or 5% of the world's population. The Rights of Persons With Disability Act (RPWD Act) 2016, defines Hearing impairment as “(a) persons with 70 DB hearing loss in both ears within speech frequencies is considered as ‘deaf’ (b) the person with 60 DB to 70 DB hearing loss in both ears within speech frequencies is referred to as  ‘hard of hearing’ and ‘speech and language disability is considered as permanent disability due to conditions like laryngectomy or aphasia which affects any of the components of speech and language due to organic or neurological causes”. As per RPWD Act, it is the responsibility of the Government and the local authorities to ensure equal participation of persons with disabilities equally with others [1]. Because education is so important to one's quality of life, it's important to research the elements that influence this population's education. This may help the government and other stakeholders of education to formulate policies that ensure equal opportunities for all irrespective of disability [1].

According to the 2011 Disability Census, India has roughly 5 million persons with hearing loss. Hearing loss is the country's second most common disability, accounting for 18.9% of all disabled people. There is no considerable data on the literacy abilities of the deaf in India, but informal observations suggest that students with hearing impairments may have to wait several years to achieve a respectable level of reading and education and that this level is not always achieved [2]. Currently, as per available data, only very few institutes offer undergraduate programs for students with hearing impairment [2]. Students with hearing impairment have struggled academically owing to a variety of causes. Low academic performance has resulted in poor employability, low income, and substandard quality of life in India, hence research into this performance and the causes associated is urgently needed [2]. Many studies in the field of deaf education were undertaken in the late 1980s and early 1990s, and society's attitude toward this sector improved [3]. Students with hearing impairments performed poorly in comparison to their hearing peers, despite numerous advancements. Students with hearing impairments lacked the necessary information and abilities to maximize their potential [4]. There were no suitable pedagogies that can be used by teachers in the classroom with students with special needs [5].

In India, schools and colleges follow general conventional teaching methodologies to teach students with deafness. The lack of expertise in pedagogy adds to the barriers faced by students with deafness receiving effective educational services [6]. Prior academic achievements, student background attributes, and eLearning attributes are considered as the factors that affect the academics of a student [5]. For a student with hearing impairment, the attributes related to hearing impairment may also need to be considered. The identification of these characteristics that influence students with hearing impairments' academic performance may aid educational stakeholders in developing tailored and individualized teaching plans for them, as well as policies that will enable them to pursue higher education possibilities. Additional accommodations are required for including students with hearing impairment in mainstream schools. The identification of characteristics that influence students with hearing impairments' academic progress may aid in the identification of such adjustments.

The analysis of factors that affect the academic performance of students with hearing impairment can significantly improve their quality of life, with technological advancements and machine learning. This can help the authorities to provide equal opportunities to all without any barriers.

The non-availability of an open dataset that comprises the details of students with hearing impairment leads to the development of a dataset. The dataset is developed by collecting data from the National Institute of Speech and Hearing, Kerala, India. This is one of the prominent institutions in the country that offer specially designed undergraduate programs for deaf and hard-of-hearing students. The data of 224 students are collected. The limited higher education opportunity for students with hearing impairment justifies the small size of the dataset. The previous research identified previous study achievement as one of the predictors of student achievement [15]. There is literature that observes that socioeconomic features can affect the performance of students [12]. The major challenge faced by students with hearing impairment is in communication and language skills. The related features are identified that include communication mode, speech therapy received, type of schooling, degree, type, and percentage of hearing loss. Under each of the three broad classifications- Socioeconomic features, prior academic features, and deafness-related features independent features are identified.

Based on the characteristics, the features are divided into 3 categories – (a) background details which include independent features like age, gender, family background, education of parents, annual income, state, and district (b) prior academic details like independent features - marks scored for class 10 and class 12, semester 1 marks for undergraduate course (c) deafness related independent features like percentage of hearing loss, type and degree of hearing loss, family history, mode of communication, Type of school, speech training, Cochlear Implant, Hearing aid use. Figure 1 illustrates the key characteristics identified which are linked to students with hearing impairments' academic achievement.

1.jpg

Figure 1. Factors related to the academic performance of students with hearing impairment

The data was collected through Google form and from the academic records of students from the institution. The workflow is represented in Figure 2. The data preprocessing was done by cleaning the data and checking the missing values. The preprocessing of data is very crucial in data analytics. Several methods deal with missing data. Some of them are ignoring the tuple, filling the missing value manually, use a global constant or attribute mean or most probable value for the missing values [16]. Considering the limited opportunity for students with hearing impairment, the dataset size is small. It is found that almost all the columns did not have missing values. Only two columns, Percentage_class10 and Percentage_class12 have 5 and 6 missing values respectively. As the dataset size is small, ignoring the tuple is not considered; instead, they were imputed by the mean of the column values. The dataset included categorical and continuous values.21 features were identified which include socio-economical, academic, and deafness-related features. Centering and scaling are performed independently on each feature by computing the relevant statistics on the training set samples. The mean and standard deviation are then saved for later use with transformation on other data. Many machine learning estimators require dataset standardization: they may behave poorly if the individual features do not resemble standard normally distributed data.

2.jpg

Figure 2. Workflow diagram

The StandardScalar() function in Sklearn library is used for standardizing the numerical features. The standard score of a sample x is calculated as:

z=(x-u)/s        (1)

where, u is the mean of the training samples, and s is the standard deviation of the training samples. It is observed that there is no change in distribution after normalization. The students were grouped into three based on their first semester marks as A, B, C -High, Middle, and low-level respectively. The Label encoding was done to convert categorical values to numerical values. The distributions of variables in different groups were found and the exploratory data analysis was done with graphical representations to obtain valuable insights into the data. The numerical features were standardized and the correlation of these features with the grade was found using correlation. The correlation matrix is plotted between the continuous attributes and the semester one marks... The chi-square test is a test for independence that compares two categorical values in the contingency table to see whether they are related. The formula of chi-square statistic used in the chi-square test is:

$\chi_c^2=\sum \frac{\left(O_i-E_I\right)^2}{\mathrm{Ei}}$          (2)

where, c is the degree of freedom, Oi is the observed value and Ei is the expected value.

Statistical significance is often referred to as probability value or p-value. A small p-value means that your data are unlikely under some null hypothesis. A somewhat arbitrary convention is to reject the null hypothesis if p<0.05. In this study, the chi-square test is used to test which categorical features are related to Grade. The null hypothesis –H0, is taken as the variables are independent of Grade. To conclude the hypothesis with 95% confidence, the p-value should be less than 0.05, which is the alpha level associated with 95% confidence level. The p-value less than 0.05 represent the relation between the grade and the selected categorical variable, that is, the null hypothesis is rejected.

The dataset is modeled with six classification algorithms- Logistic Regression, Decision Tree, Support Vector Machine, KNN, Random Forest, and Naïve Bayes. The performance of models is compared and it was found that the performance of Random Forest is better than the others. The model is evaluated using classification report and values -accuracy, F1-Score, precision, and recall. The k-fold cross-validation technique is used to improve the performance of models.