MSIT Theses and Dissertations (2017)

Permanent URI for this collection

http://hdl.handle.net/11071/5625

Browse

Browsing MSIT Theses and Dissertations (2017) by Subject "Artificial Neural Networks"

Now showing 1 - 2 of 2
  • Thumbnail Image
    Item
    A Model for the classification of student neediness using artificial neural networks
    (Strathmore University, 2017) Manyasi, Eunice Engefu
    Financial aid has been used worldwide to assist students at higher learning institutions finance their education. The aid has majorly been offered by the government, private companies and non-governmental institutions in form of loans, grants, scholarships and work study programs. It has made great progress in increasing the enrolment rate of students to higher learning institutions. The aid is usually given to applicants who have been selected after applying for the aid, and a committee ensuring that they have meet the set criteria to be awarded. Currently the number of applicants applying for financial aid has increased leading to challenges of errors and bias in the selection and award process due to overwhelming data which becomes too complex for the committees to analyse. This has led to some more deserving students not receiving the financial aid due to inaccuracies. Artificial intelligence has been applied in various fields for the analysis and classification of huge amounts of data. It has been applied in finance to predict the credit rating of customers which uses a similar concept in classification of applicants. The research sought to apply machine learning to in the selection and award process of needy students. Historical financial aid data which was labelled as awarded and not awarded, was used to train the feed forward neural network learning model. The inputs used included parents occupation and income, family income and family spending. The research employed experimental research to determine the variables that best identified the needy students and qualitative research to get the ideas and opinions of participants with regards to the study. The model accurately classified 2955 instances as true positives and 18 instances as true negative out of 3043 instances, giving it a 97.6% accuracy.
  • Thumbnail Image
    Item
    Sound analysis prototype to enhance physical security in academic institutions
    (Strathmore University, 2017) Omondi, Kevin Ochieng’
    Competence in the provision of security to the civilians in Kenya has generally deteriorated and hence negatively affecting the public trust accorded to security agencies. Indeed, the police to civilian ratio is low and this has affected the institutions of learning as they have become new attack grounds for the terrorists. Institutions of learning have suffered the worst since they are expected to be accountable of their own security in many cases. As a result, many institutions of learning use available security agencies, most of which employ outdated and less efficient means of implementing security. Examples of commonly used physical security techniques include the use of security guards, perimeter walls, some places use turnstiles as well as CCTVs. The inefficiencies that comes along with these security measures has still however exposed these institutions to great dangers of insecurity. This study proposes the use of sound classification to enhance physical security. The solution relies on the integration of the possible solutions of the artificial neural networks (ANN) in sound classification to detect sound variations in the leaning institutions. It is expected that decisions made through classification assist security personnel on the ground to tighten the physical security. The solution offers automatic analysis of the recorded sound from the environment, compares it to the stored dataset which has urban sounds and the score labels displayed on the output screens for the security personnel to help them enhance the available physical security. The usage of scientific research methodology through experimentation ensured that the sounds were captured, the dataset sounds were collected and trained for comparison to take place and finally results validated to prove the theory. The system proved an accuracy percentage of 78%, and the efficiency, user friendliness and reliability were al passed.