MSIT Theses and Dissertations (2021)

Permanent URI for this collection

http://hdl.handle.net/11071/12748

Browse

Now showing 1 - 20 of 20

Twitter sentiment analysis tool for detecting crime hotspots: a case of Nairobi, Kenya
(Strathmore University, 2021) Onyango, Kevin Omondi
Insecurity brought about by crime continues to be a major thorn in the flesh of citizens leaving in Kenya's urban centres. Although, incidents of crime are frequently reported from different regions in Kenya, there are more cases being reported from Kenya's urban centres than the rural ones, especially the informal settlement areas. This has been attributed to the rapid urbanisation of Kenya’s major towns. Violent crimes are costly. Murders, rapes, assaults, and robberies impose concrete economic costs on the victims who survive as well as the families of those who lose their lives, in the loss of earnings and their physical and emotional tolls. The Kenyan government has invested heavily in setting up Internet Protocol Circuit Television Cameras (IP CCTVs) in Nairobi's Central Business District in a bid to curb crime. In 2015, the government implemented a community policing strategy at various levels namely, market, estate, house level among others. This community policing is known as Nyumba Kumi. The culture in urban centres especially Nairobi makes it very difficult to implement Nyumaba Kumi. For instance, Nairobi is a city where people are less concerned with the affairs of their neighbours. For Nyumba Kumi to be effective in Nairobi, a culture change has to occur. Culture changes usually take time. On the other hand, CCTVs have proven to be a useful tool in tracking down criminals and bringing them to book. However, maintaining CCTVs is quite expensive and CCTV footages have been reported missing in some cases whenever investigators needed them. Data mining algorithms can be employed to fetch useful patterns on Social Media posts especially Tweets from Twitter to monitor crime. This study proposed a Twitter sentiment analysis tool which was used to detect crime hotspots in Nairobi. The tool employed machine learning techniques to a build binary classifier in detecting crime hotspots. This research fetched sample crime relevant tweets from Twitter which were used to build the corpora. Then a Support Vector Machine model was trained and validated based on the labelled text data using bigram features and term frequency-inverse document frequency weighting. In order to determine what combination of features provided the most desirable performance outcome on the data collected, the SVM model was compared to Naive Bayes, K-nearest neighbour and Random forest machine learning algorithms. Based on the results from the experiments, it was found that the best way to create a model for detecting crime hotspots using Twitter is the use of an SVM machine learning algorithm with bigram features weighted using tf-idf. The SVM model produced an accuracy of 88% making it the most accurate compared to the rest.
Navigation guidance to pedestrians’ road accident safety in Kenya using Bayesian analysis
(Strathmore University, 2021) Kuria, Paul Munene
In most countries, policy makers find the safety of pedestrians to be a major concern because many pedestrians are involved in road traffic accidents. Roughly, 1.35 million people's lives every year are shuttered because of road traffic accident, among these fatalities pedestrians are most vulnerable road users given that the road infrastructure is designed with less or no consideration to safety of a pedestrian. Therefore, the purpose of this research was to propose a navigation aid system that informs pedestrians about the safety status of a given road route in Kenya. Experimental research was used because it facilitates the manipulation of variables which is essential in this study. In designing the system, the research implemented the Agile development methodology. The data used was from the secondary references which included traffic accident fatalities records from NTSA. The data was used to train a Bayesian Linear Regression model, which predicts the pedestrian’s road accident fatality likelihood in a given road route. The model’s results, posterior distribution likelihood, are presented in terms of color coding on the Google Maps routes via a mobile application, where red indicates high danger, red for high-medium, brown for medium, yellow for low-medium, green for low. The study found that pedestrians constitute 54.14% of total road accidents fatalities and pedestrian’s risk of being involved in a fatal road accident is influenced by following factors, population increase with dependency of region urbanization, the time of the day with 2000hrs-21000hrs posing highest risk, tendency of using same road route and gender in which a male’s risk is 73% higher in comparison to female’s risk. These findings were used in fine tuning the Bayesian model and enriching the system’s web portal reports. The implication of the research is that a pedestrian is able to know the safety status of a given road route in respect to road accidents via a mobile application enabling him/her to choose among the available routes and taking the necessary safety precautions based on the safety status of the chosen route.
Food recommender system for Diabetes Type 2 patients
(Strathmore University, 2021) Kariuki, Esther Muringo
Diabetes mellitus is a chronic medical disorder that arises when the body does not produce enough insulin or when the body does not utilize the insulin produced effectively. Insulin is the hormone that regulates blood sugar in our bodies, when blood sugar is not regulated it leads to hyperglycemia which is excess blood sugar levels or hypoglycemia which is very low blood sugar levels. Due to urbanization that comes with very busy day to day schedules patients tend to pay little or no attention to their eating patterns leading them to opt for quick fixes such as fast foods, less balanced diets, and intake of a lot of processed food. The number of cases of diabetes patients has been steadily incrementing over the past few decades. Lack of proper diabetes management results in long-standing complications that end up using up on an individual’s resources such as money and time spent seeking medical attention now and then. Diabetes management and control of blood sugar levels are usually done through pharmacotherapy which is the use of medication alongside nutritional therapy which involves eating healthy diets. For nutrition therapy to be effective, patients must consume nutrient-dense diets foods. Patients should take caution on their carbohydrate’s intake, glycemic index, and glycemic load levels of foods they consume, this way they can control and maintain the blood glucose levels close to normal. Today, with the tremendous growth in technology we see an increase in the adoption and use of health recommender systems that are slowly becoming a close companion to an individual. A health recommender system can study the user, gather relevant information and recommend what suits the user best, hence making life easy. This study sought to develop a food recommender system expressly for diabetes Type 2 patients which will incorporate the use of a glucometer, a medical device used to assist patients and the caregivers monitor blood glucose levels and use that data to adjust nutritional therapy. Based on their sugar levels the recommender system will advise the patient on the appropriate foods they can consume at that time, ensuring the blood glucose target is attained hence reducing chances of sudden blood spikes and dips. Once the patient keys in the food they want to eat the model will respond by giving the patient the go-ahead to consume the food. In this study, we tested the Naïve Bayes algorithm with collaborative filtering to recommend food and achieved a prediction accuracy of 90.0%. The algorithm outperformed decision trees which gave an accuracy of 78% and the Support vector machine which had an accuracy of 75%.
A Blockchain-based distributed hybrid system for tracking power distribution in electrical power systems
(Strathmore University, 2021) Kimiti, Raphael Leikari
Power distribution systems carry power the last few miles from transmission or sub-transmission to consumers through wires either on poles or underground. These systems encounter perpetual issues of power losses, and they have little or no real-time monitoring of power flow. These challenges represent a significant challenge of electricity use. The aim of this research is to develop a blockchain-based distributed hybrid system for real-time tracking of power distribution in electrical power systems. The study devised a solution that can aid in power re-routing to end-users in the distribution grid in case of transformer failure. The solution consists of a blockchain-based platform to help re-route power in the cases of any power outage and blackout and relaying the relevant information regarding the cause of the disturbance to the utility center in real-time. The rapid application development methodology was used for this study because it offers a quick and flexible way to discover and validate the idea and facilitates easy incorporation of changes to the prototype. The results of this study show that continuous power availability to end-users is achieved through the application that has been developed using blockchain. The proposed solution can enable industries, residences, and other end-users to benefit from having reliable and uninterrupted power flow. In conclusion, it is expected that the proposed solution will improve grid access which in turn can eliminate unpredictable outages. By using the proposed solution, power distribution companies can easily monitor the grid and have quality supplies to end-users.
A Computer vision-based model for crop yield prediction using remote sensing data
(Strathmore University, 2021) Kiragu, Daniel Mburu
Arguably, crop yield data forms the most important measure of crop productivity in agriculture. With adequate crop yield data, local and international bodies can develop effective agricultural policy leading up to sustainable food supplies and elevated food security. However, timely acquisition of crop yield data can be a cumbersome task as existing crop yield prediction approaches face numerous challenges. In this study, these challenges are identified as high cost and high dimensionality of data required for the prediction activities as well as limited scaling of the resultant prediction models. In efforts of overcoming these challenges, this study leveraged an alternative source of data to design and develop a cheap, accurate and scalable deep learning model using convolutional neural networks. Satellite imagery datasets were used as the primary and only source of data for training the model. This benefited the study in two major ways. Firstly, off, the approach automatically took care of the high dimensionality problem as demonstrated in the GEMS data. Second, satellite imagery data is readily available globally, a factor that greatly reduced the costs needed to collect real-time data for the study. Validation of the developed model was done using 10% of the overall dataset acquired. Reliability of the model in performing crop yield predictions was captured using an MSE loss function for each epoch trained. Cumulatively, the model achieved an MSE loss score of 3.6.
A Deep normalized neural network model for strawberry fungal leaf disease detection
(Strathmore University, 2021) Kerre, Deperias Webula
Strawberry is one of the cash crops that are being grown in Kenya for both export and local consumption. However, strawberry fungal leaf diseases are threatening the existence of this crop which is an important input in the agricultural production sector. The types of strawberry fungal leaf diseases resulting to greater losses in production include Strawberry Leaf scorch, Strawberry Leaf Spot and Strawberry Leaf Blight. The biggest challenge the farmers face is that of correctly classifying these diseases based on observable leaf features. Famers have incurred losses due to poor/incorrect control measures which results from the misdiagnosis of these diseases. This scenario is more pronounced in rural settings where the farmers have a limited access to expertise in modern agricultural production. As a result of this, automated classification of strawberry plant fungal leaf diseases is highly desired. The literature review found several computer vision techniques that have been leveraged in Strawberry fungal leaf disease detection. Among these solutions are the convolutional neural network-based models. Despite the high detection accuracy, the models do not cover another of strawberry fungal leaf diseases such as leaf spot and fail to generalize well on unseen data. The models also do not consider cases where more than one disease occur on the same part of the plant, in this case the leaf. In this study, a deep learning model was proposed for classifying fungal leaf diseases in strawberry based on an experimental research design. The model generalized well on previously unseen data and considered a scenario where multiple diseases occur on the same leaf (Leaf Scorch and Leaf Blight). The model also covered strawberry Leaf Spot that was not covered by any of the existing deep learning models. Data samples containing a total of 1,134 leaf images, categorized into five classes including healthy leaf images were split into 80% training and 20% validation. The disease classes include strawberry leaf spot, leaf scorch, leaf blight and a class where two diseases (Leaf Blight and leaf Scorch) occur together. The model was trained on 30 epochs from scratch with batch normalization implemented within the convolutions in Keras framework and validated using a confusion matrix. The model achieved an outstanding classification accuracy of 98%, precision of 97% , recall of 95.7% and an F1-score of 96.3%.
A Prototype for profit maximization using Apriori data mining algorithm: case of Kula Kona restaurant
(Strathmore University, 2021) Okeyo, Seth Ouma
Most businesses use different promotional and pricing methods to improve profits, revenues, and sales volumes. For example, a restaurant manager may change prices to encourage sales of food items. Also, he or she may in a special way advertise or present advertise the items to increase customers’ awareness and demand. This has become cumbersome and has made the management of such businesses difficult and has informed the decisions to develop systems with aims of coming up with a solution to this, most of the systems are complicated in nature and difficult for the users to apply, most of them are also rigid to platforms/system requirements since they were developed in old and probably un-scalable platforms. The aim of this research is to formulate a prototype for profit maximization using Apriori data mining algorithm. This is achieved by applying the algorithm on existing sales knowledge bases with other given parameters, some kept constant and others varying, and the algorithm is able to determine the sales patterns using different internal and external parameters. The prototype then automatically analyzes the patterns and come up with reports and summaries which can aid in decision making and consequently profit maximization with the optimal prices of the goods, which is advantageous to both the restaurant owners and clients. The research site is Kula Kona restaurant located in the Hurlingham area within Nairobi. The research design is used to conduct the scientific study and descriptive approach to demonstrate the effects of adjustments of different variables which help understand the behavior and effects on other variables in relation to sales at a given time. The Data-driven modelling methodology was used in this model development. The methodology was ideal since it relied on retrospective data and it performed at the accuracy level of 93.71% and a mean square error of 0.039. The results were great and showed that Apriori algorithm is the best fit for this type of machine learning prototype.
A Framework for secure medical records: a case study of Kenyatta National Hospital
(Strathmore University, 2021) Otieno, Theodulus Odhiambo
Healthcare information systems are largely viewed as the single most important factor in improving healthcare quality and reducing related costs. However, managing Information Security is becoming more challenging because of security incidents due to non-compliance by health workers. This was an intrinsic case study to gain a better understanding of how a medical institution can embed information security culture in the management of security of its medical records. The application of case study research is appropriate in a new and emerging area of research as it a strategy that allows for an in-depth exploration of the phenomenon. A survey questionnaire was given to the employees of the Ear Nose and Throat department of Kenyatta National Hospital to measure the human aspects of the Information Security Program. Interviews were used to further explore the perceptions of respondents and probe for more information and clarification of answers. The study shows that management support, training and awareness, well-articulated and visible security policies will have a significant positive effect on compliance and hence the security of medical information. Additionally, the study showed that the employees have a great sense of commitment towards protecting the information of the organisation. This is because the management has taken the initiative to lead by example, avoids punishing workers for non-compliant behaviour and motivates the employees towards a security-conscious behaviour. This study sought to explore how the human factor may influence information security and how this can be harnessed together with technology to improve the security of medical records.
An Automated personality classification based system for assisting in choosing a career using data mining techniques
(Strathmore University, 2021) Josiah, Susan
For successful career development in today’s world of work, the empowerment of individuals as autonomous decision-makers is fundamental. This empowerment aims to help individuals in the acquisition of decision-making skills when making transition decisions. A lack of self-awareness is a contributing factor as to why people land in the wrong career. After in-depth research, the researcher found out that deliberating individuals encounter countless challenges in the process of career decision making. After establishing that one’s personality can influence how one performs in the work place depending on the career they are in, the researcher sought to create self-awareness to individuals faced with the dilemma of choosing a career that they can thrive in best. To achieve this, the researcher has developed a web application that can automatically classify a person’s personality and recommend a career that fits their personality. To achieve its purpose, this study assumed a purposive sampling technique that drew at least 70 respondents based on two classes of participants. One class was composed of high school students and persons considering changing careers. To develop the web application, the researcher used Web Development Life Cycle (WDLC) methodology which combines the components of both Systems Development Life Cycle (SDLC) and Prototyping. WDLC contains two iterative steps of graphical development and functional development. This methodology was found efficient in reducing development time, accords more structure to the research problem and ensures user involvement throughout the development life cycle. From the study findings, it is evident that there is a strong relationship between personality and career choice and that a career recommendation system based on one’s personality can make career decision process manageable. This system will be helpful to learning institutions when advising students on career paths to pursue based on their personality type. The prediction of personality is based on MBTI 16 personality types and the data mining algorithm used for classification is K- Nearest Neighbour.
Mobile application for early detection of Malaria in children: case of Western Kenya
(Strathmore University, 2021) Macharia, Georgina Wangeci
Malaria is the most infectious disease and continues to be a major global health problem with part of the world’s population being at risk to various degrees of malaria risk. In many endemic countries, the clinical diagnosis has been proven to be the only method used to decide on the correct treatment even though the method is not that accurate and may be limited by the low specificity of the various signs and symptoms of malaria. Some of the challenges affecting the early detection of malaria include and are not limited to severe anaemic and respiratory diseases in children and delayed detection of malaria leading to irreversible and fatal complications in children. These challenges led to the implementation of a mobile application for early detection of malaria in children. Several measures have been made in combating malaria, however, the indicators in Africa still do not show any promise for elimination in the future as the infections still result in high mortality and a rise in the high rate of children affected with malaria. Reducing the number of deaths in children affected by malaria would yield huge gains in reducing the overall under five mortality and morbidity rates in malarial dominant areas. The purpose of this research is to be able to detect malaria at an early stage in children in the regions of Western Kenya. The solution proposed, was coming up with a mobile application that will facilitate the most suitable and convenient way of malaria disease detection, especially in rural and remote regions. The camera of the smartphone will act as a microscope and there will be no need to attach it to the eyepiece of the microscope. This enhances mobility and the Remote Health Worker is able to diagnose the patient and offer treatment. This will allow real time treatment and the records will be uploaded to the database for the next visit from the Remote Health Worker. The study used agile software development model to design, develop and test the application since it is iterative. The mobile malaria detection application was developed and tested to be used by the resulting model which had an accuracy level of 94%. The findings from the usability acceptance test showed that the users acknowledged that the application was easy to navigate, use and the instructions were clear to use as a first-time user.
A Classification model leveraging Electronic Immunization Records to predict child immunization completion: case study - Mukono Health facility
(Strathmore University, 2021) Kembabazi, Bertha
Immunisation is one of the most cost-effective public health interventions, it prevents child deaths by strengthening immune response and preventing diseases that are not only deadly but also easily transmitted. However, countries like Uganda still face challenges that limit the attainment of immunisation completion targets like late and missed doses. It is noted that many children who start immunization do not follow through to the last dose which leads to incomplete doses hence no full protection and also missing some vaccinations which protect the child against other diseases, this in the long run exposes the child to the risk of contracting deadly diseases as well as spreading the same to others. There is potential to use data from electronic immunisation records systems to get projection insight to follow up on participants to increase access to immunisation. This study uses a random forest classification algorithm to develop a model to predict completion rates of infant immunisation to improve immunisation service delivery and utilization. This model predicts those likely to complete the recommended immunisation vaccines as per the schedule using DPT3 as an identifier classified into three categories. The categories were coded as 3 for those likely to miss, 1 for those who will receive on-time and 2 for those who are likely to receive the scheduled vaccine late. Using existing secondary electronic immunisation records data from the MyChild System implemented at Mukono district health facility, the data used was collected between 2015 and 2020. 75% of the data was used as training data while the other 25% was used as test data for the model. The predictors of this model include child dates of vaccine dose administration, the exposure to tetanus, whether a child was exposed to HIV, the date of birth and whether the caregiver was counselled. The model was tested and validated to give accurate predictions and the measure of accuracy as an output.
A Differential diagnostic tool for obstructive lung diseases in adults using classification models
(Strathmore University, 2021) Ochieng, Daphne Faith
The non-specificity of affective symptoms between asthma and chronic obstructive pulmonary disease is a key challenge in medical practice. This is even more complicated when physicians fail to recognize the co-existence of these two diseases, a condition known as asthma-chronic obstructive pulmonary disease overlap. This occurrence mainly affects smokers, middle-aged and older age groups. Failure to diagnose a disease correctly and in a timely manner often leads to administration of wrong drug therapy, delayed treatment or wasted financial resources. To reduce the risk of misclassification, a differential diagnostic tool that can differentiate among patients with asthma, chronic obstructive pulmonary disease and asthma-chronic obstructive pulmonary disease overlap was developed based on measurements of spirometry, blood eosinophil count and smoking history. The study employed a judgmental sampling technique to draw 184 samples from the National Health and Nutrition Examination Survey (NHANES) database based on three classes of obstructive lung diseases. Rapid application development methodology was used as the software methodology upon which the architecture was designed and developed. A comparative analysis was made between a number of classification algorithms including K-nearest neighbour, support vector machines, logistic regression, multilayer perceptron and random forest. The results demonstrate that the differential diagnosic tool can correctly classify patients with obstructive lung diseases with an accuracy of 93.94%, showing an automated approach that would aid physicians in making preliminary diagnoses, leading to optimization of time resources, lower medical device costs and better patient outcomes. Further research regarding the tool’s improvement should focus on using a more robust dataset and evaluation in liaison with a physician in a real clinical setting.
A Model for predicting pre-delinquency of credit card accounts using Extreme Gradient boosting
(Strathmore University, 2021) Kisengese, Antony Mwawuganga
Credit risk is one of the significant risks that financial institutions that advance credit in credit cards are exposed to. Credit card accounts are usually classified as “good” or “bad” depending on the propensity of the cardholder to settle their debt on time. The latter usually pose a significant negative impact to the issuer’s books when the credit card account falls into late collections and recoveries are futile resulting to bad debts. Ensemble classifier algorithms have demonstrated greater performance in classification and regression problems due to their ability to trade-off bias and variance factors. In this study, an Extreme Gradient Boosting ensemble classifier was implemented based on cardholder personal characteristics and transaction patterns with the aim to minimize defaults in the late collection stages by identifying credit card accounts that exhibit early signs of delinquency way before the cardholder misses payments. A credit card dataset from the UCI Machine Learning repository was used to train and validate the model, which achieved a prediction accuracy of 81.62% and outperformed a set of single classifiers that were used in benchmarking. Depending on each score, the issuer will make informed decisions of how well to proactively engage the cardholder to identify the best way of intervening in their financial situation and mitigate the risk of missing payments.
IoT pulse oximetry model for early detection of COVID-19
(Strathmore University, 2021) Bonyo, Lesley
Coronavirus disease 2019 (COVID-19), which was declared a pandemic by the World Health Organization, is a respiratory illness caused by severe acute respiratory syndrome coronavirus-2 (SARS-CoV-2). With no specific treatment against SARS-CoV-2, early detection of COVID-19 is vital to effective tracking and management of the disease. For this reason, several diagnostic strategies have been implemented to identify COVID-19 infection, to test for past infection and immune response. These include molecular tests such as RT-PCR, antibody tests and medical image analysis. While the RT-PCR is the gold standard test for confirming the COVID-19 infection, it requires specialized labs and is time consuming. As an alternative, Chest X-Ray and CT images using deep learning algorithms have been used. However, because of harmful radiation doses these approaches cannot be relied on for patients’ screening. Hence, there is a need for a less expensive, more accessible, and faster detection model to identify COVID-19 disease. Physiological data such as temperature and oxygen saturation can aid in COVID-19 detection and monitoring of COVID-19 patients. The symptoms for a person who is indicative of Covid-19 include shortness of breath, abnormal heartbeat, and abnormalities in lung function like the symptoms of pneumonia. Further, there is a target oxygen saturation range for patients with COVID-19 recommended by the National Institutes of Health. Such data can be continuously collected to monitor health of individuals using Internet of Things (IoT). The research tested various machine learning algorithms and implemented a low cost IoT based system with a KNN model which produced the best results. The KNN model, based on monitored oxygen saturation levels, heart rate and other COVID-19 symptoms, made predictions of person’s health based on their possibilities for COVID-19 infection an accuracy of 66.67 percent. This can aid in early detection of COVID-19 symptoms to influence early testing of individuals and to assist hospitals in remote monitoring of symptoms in patients who have contracted the virus.
A Honeypot based malware analysis tool for SACCOs in Kenya
(Strathmore University, 2021) Mwendwa, Keith Mwesigwa
Kenya had her first established Savings and Credit Co-operative (SACCO) society in 1908 and to date, the SACCO societies have grown into a Billion-dollar industry. SACCOs contribute 5.72% to Kenya’s Gross Domestic Product (GDP) and are significantly changing the lives of Kenyans in almost all sectors of the economy. Like other sectors, SACCOs are facing growing cyber threats that have potential to affect their performance. The report by Serianu of 2018 indicates that SACCOs have poor visibility on enterprise cybersecurity and thus are poorly prepared to anticipate risk, detect vulnerabilities, respond to incidents and contain threats. Further, SACCOs have low budget allocations and inadequate skilled staff to advise in prevention and protection against threats. Because of this, SACCOs across the globe lose hundreds of millions of dollars annually. The Serianu Cyber Security Report of 2018, indicates that the global cost of cybercrime was at 600 billion dollars in 2015, which had risen by $100 billion from the previous year. The report indicated the SACCOs were the most affected, while the affected organizations lost money, experienced downtimes and reputation damage. It is observed that many SACCOs in Kenya are slowly putting up measures to prevent, detect, and remediate cyber-attacks with minimal resources. This study intends to help SACCOs have a paradigm shift in how to detect and respond to malware by developing a prototype. The literature review brought to light the different applications of honeypot solutions, but the solution is not common within the SACCO industry. The prototype, a honeypot that was used for malware analysis in order to determine breach scenarios and common cyberattacks showed outstanding performance when run for a few days, in capturing malware, and helping in their analysis. The proposed solution enables SACCOs to better mitigate and possibly reverse Cyber-attacks on their infrastructure due to the information they get from analysing malware. Development of the prototype was based on Rapid Application Development methodology to build a robust malware analysis tool on Honeypots and was tested for reliability where it showed an outstanding accuracy level as all attack traffic was captured and logged. While from the first 24 hours of uptime, in 100 captured attacks, the prototype was able to give Md5 hashes of 11 malwares, the prototype captured the IP addresses associated with the rest of the attacks which can be blacklisted by a SACCO employing this tool.
A Customer segmentation model using logistic regression: a case of Telkom Kenya
(Strathmore University, 2021) Omonge, Jevans
Market segmentation is a marketing strategy that has been widely used by many companies globally. With the ever increasing volume of client data, many companies are now unable to clearly cluster their clients into their respective segments, subsequently providing them with products and services that are best suited for them. Telkom Kenya is currently the third largest telecommunications company in Kenya. Currently, telecommunications companies do not have well defined marketing plans for their customers based on their daily expenditure. Some companies, for instance, may provide their customers with additional voice airtime even when such customers spend significant amounts of credit on data bundles rather than the actual voice airtime. One way of overcoming this challenge is by enhancing the current state of market segmentation in telecommunication companies in general. In this study, we present an approach that incorporates business intelligence, big data and machine learning in order to achieve customer segmentation. The study is based on data collected from the spending patterns of Telkom Kenya customers. When designing the customer segmentation model, the fundamental steps in the designing of any machine learning model were followed. To begin with, data was collected from the CRM department of the company. Key trends and inferences from the data were obtained from extensive data visualization that was performed on the data. The data was then formatted to ensure that it was consistent before performing feature engineering with the primary purpose of improving the quality of the features. Thereafter, the data was split into training and testing sets. Finally, the processed data was fed into the actual machine learning models. The main classification algorithms evaluated in this study are Logistic Regression Classifier, Linear Discriminant Analysis, K-Nearest Neighbor, Decision Tree Classifier and the Guassian Naive Bayes Classifier. Of the five, Logistic Regression Classifier was found to have the cross-validation accuracy and was thus embraced for the customer segmentation process. The results of this study therefore show yet another potential application of machine learning in marketing in general through customer segmentation. As seen from the results, the machine learning has been able to categorize customers into their respective categories with 71% accuracy. Through the classification, Telkom Kenya is in a position of marketing their products and services to the right group of customers, thereby ensuring that their marketing strategies are effective.
A Model to measure online student engagement using eye tracking and body movement analysis
(Strathmore University, 2021) Mido, Jude Austin
Many institutions are adopting remote learning as way of expanding and offering their programs to mostly undergraduate students and adults seeking further education or training, and as a way of doing this at low costs; without constructing new buildings. However, measuring student engagement so as to focus attention on students who are struggling and manage students in a large class presents an extra challenge to teachers, when they have to do it virtually on eLearning platforms. The focus of this study was to build a model to measure student engagement using eye tracking and body movement analysis through web cameras, to help in tracking student engagement. The proposed solution is aimed at assisting in maintaining student engagement during remote classes, as it would be in a traditional classroom, and to enhance the learning process for both the student and the teacher. The proposed solution aimed to achieve this using computer vision algorithms; using computer vision to track the eyes and detect the hand, then analyze the movements to provide indices used to measure the engagement level. The model was tested using a prototype that analyzed recorded videos of students attending a remote class. The model achieved an accuracy of above 80%.
An Algorithm for identification of terror events and hotspots using K-means and discriminant analysis approach
(Strathmore University, 2021) Ndambuki, John Kelvin
This study aimed at developing an algorithm for the identification of terror events and hotspots using K-means clustering and discriminant analysis, with pre-terror recruitment, planning and preparatory activities as the determinant factors. Rules and logic for quantifying the risk state of a pre-terror event based on the values of the constituent determinants of that pre-terror event, for example recruitment as determined by factors such as age of recruits, location and terror organization involved, were developed. Kmeans clustering was used to come up with two clusters based on the risk value combination of the pre-terror event activities. The two clusters represented the outcome of having a terrorist event happening and a terrorist event not happening, and were duly labelled. Discriminant analysis was used on the now labelled clustered dataset to come up with two identification functions, one for terror event happening and another for a terror event not happening. Unseen possible values for pre-terror event activities were fed into the developed algorithm and an identification of whether a terror event would take place and possibly where was accomplished. The purpose of this identification as per the aim of the study was to offer insights to the organizations dealing with counterterrorism activities. The general public would benefit from this effort once a possible terror attack was prevented before it actually took place. The main research question that guided this study was how accurately a possible terrorist attack incident could be identified before it happened. This study used data from the Global Terrorism Database (GTD) retrieved from the National Consortium for the Study of Terrorism And Responses of Terrorism (START) to come up with geospatial datasets as part of a geodatabase with spatial and temporal information on areas that have been attacked before and the risk values of their constituent pre-terror events.
A Blockchain-based prototype for cybersecurity threat intelligence sharing: a case of Kenyan banking and insurance financial institutions
(Strathmore University, 2021) Kibuci, Wanjohi Stephen
Cybersecurity threats to financial institutions have become more sophisticated and challenging to deal with. The growing dependence of financial institutions on cyberspace makes cybersecurity preparedness against threats important to achieve a financial institution's mission and vision. In this context, cybersecurity preparedness is the process in which a financial institution can protect against, prevent, mitigate, respond to, and recover from cyber threats. Traditionally, most organizations share threat intelligence through ad hoc methods such as emails and phone calls but there is a need to automate threat intelligence sharing where possible to improve cybersecurity preparedness. To address this issue, and enhance cybersecurity and trust, a blockchain-based approach can be employed to share threat intelligence. This study aims to leverage blockchain technology by developing a prototype to automate cybersecurity threat intelligence sharing in financial institutions. The study used a quantitative approach in data collection using structured online questionnaires with close-ended questions and open source datasets and data analysis using several analytic tools. The prototype has been developed using the Rapid Application Development software development methodology using open-source Oracle Virtual Box that runs on Linux Operating System
An IoT prototype to mitigate human-wildlife conflict
(Strathmore University, 2021) Thuo, Ng'ang'a Thuo
Due to the accelerated population growth in Kenya and around the world, pressure on and competition for resources in game parks, ranches, conservancies and game reserves has grown tremendously. This has led to the reduction of wildlife in terms of population due to poaching and territory invasions by both domestic animals and human beings. This problem calls for a technological solution to mitigate the constant conflict between wild animals and human beings. The eventual fallout from this conflict, if it were allowed to persist, would be the extinction of important flora and fauna that contributes a great deal to natural ecosystems vital for the survival of the human race, the habitability of the planet and the sustenance of economies which gain from the existence of unique wild animals and plants through tourism and scientific research into wildlife. The main objective of this research is to mitigate human-wildlife conflict by developing a prototype that combines web, loT and SMS technologies. In addition, this research also applies the Rapid Application Development methodology to develop and test a sensor-based system for monitoring animal activity in human-wildlife conflict prone areas through Internet of Things. Its physical architecture consists of an Arduino microcontroller, a Wi-Fi shield a breadboard and a motion sensor. In this architecture, the Arduino microcontroller powers the Wi-Fi shield which then connects to an available access point and in so doing, the internet. The motion sensor detects motion data and sends it to the Wi-Fi shield. The logic for this process is written and compiled into the Arduino microcontroller and Wi-Fi shield via a C++ program. The data is sent via the HTTP protocol to an A WS lambda function written in python processes this data. Once the data is processed, an SMS is triggered and sent to registered phone numbers to warn them of imminent conflict and advise on steps to take. The SMS message also helps the recipients to plan better and deploy resources in a more organised fashion to areas where conflict is rife. This solution is also low cost, accurate and can be implemented at scale along boundary areas. The results in this research show that it is possible to combine web, SMS and loT technologies to successfully mitigate and reduce human-wildlife conflict.

Browse

Browsing MSIT Theses and Dissertations (2021) by Issue Date

Results Per Page

Sort Options