MSIT Theses and Dissertations (2016)

Permanent URI for this collection


Recent Submissions

Now showing 1 - 5 of 10
  • Item
    Internet of things for monitoring environmental conditions in greenhouses: a case of Kiambu County
    (Strathmore University, 2016) Kanake, James Maina
    Efficient management of greenhouse farming is a challenge to ensure high yield production. This is a great challenge to farmers who do not have a reliable mechanism to ensure the optimum environmental conditions for their crops. Farmers are opting to look for solutions from technologies such as Machine to Machine and Internet of Things. Machine to Machine Communication refers to solutions that allow communication between devices of the same type and a specific application through wired or wireless communication networks. Moreover, Internet of Things is a connection of physical things to the internet which makes it possible to access remote data and control the physical world from a distance. These types of solutions allow end-users to capture data about events and transfer it to other devices but they do not allow broad sharing of data or connection of the devices directly to the Internet. In this thesis, the researcher investigated the use of machine to machine communication by having small electronic devices equipped with sensors that when deployed in a farm they can record the environmental conditions and communicates the information to the farmers. Moreover, the different types of crops grown in greenhouses at Kiambu County. Thereafter, the information was analyzed and sent to relevant end users such as the farmer and a metrological department that will enable them to monitor and adapt to the environmental conditions. The research used applied method of research, interviews and questionnaires to gather data. Therefore, an IoT prototype was developed to gather the critical environmental conditions in a greenhouse. The recorded data was transmitted by wireless networks using machine to machine (M2M) technologies from the sensors to the cloud platform, Intel IoT analytics dashboard, for real-time predictive analysis of the environmental parameters. An email notification was sent to alert the farmers when the parameters exceeded the threshold which were preset. This IoT prototype was used in small to large commercial indoor operations as well as small personal gardens.
  • Item
    Artificial neural network model for inflation forecasting in Kenya
    (Strathmore University, 2016) Mwangi, Carolyn Naomi Wanja
    Forecasts are important in decision making and entail prediction of a future state of a particular subject of interest. These forecasts depend heavily on historical data and the assumption that the past behaviour of forecast inputs will replicate itself in the future. Current linear and macroeconomic theory forecasting models used in Kenya lack reliable accuracy when predictors are futuristic and subject to changes over time. Artificial Neural Network (ANN) allow for the model to be more versatile in incorporating new predictors without altering the structure of the model. They work exceptionally well in environments that are nonlinear and where data is noisy and sometimes unavailable. The structure for the proposed model is a Neural Network with Back Propagation learning algorithm incorporating rainfall and M-Pesa use effects as additional inflation variables. The Backpropagation Neural Network was selected as a useful alternative due to the non-linear data used and to facilitate forecasting of future values. The adaptability of ANNs makes them most suitable for dynamic forecasting and classification problems. The results obtained from the model indicated that the back propagation was an appropriate algorithm that can be implemented in the process of inflation forecasting. The forecasting was done based on inflation variables identified as true inputs to the process of inflation forecasting. The model accuracy performance at 71.4286 % showed that the model is reliable as a tool for inflation forecasting. The study found that the optimum learning rate for the model was 0.5 while the momentum was at 0.9 for the training and 0.7 for the testing and validation data. Total iterations varied between the train, test and validate phases.
  • Item
    A prototype for tracing missing children : a case of Nairobi County
    (Strathmore University, 2016) Ndeto, Martin Ndithi
    Tracing missing children has been quite hectic for parents and care givers. A missing child is vulnerable to risks associated with drugs; poor health; involvement in criminal activities for survival, assaults, murder, rape and infection with killer diseases. Currently, there is lack of coordination in departments dealing with issues concerning children and no timeliness for the police department in handling this process. In addition, there is no convergence among the agencies involved in child protection. Several solutions have been proposed among them, the “App for the loved ones”; which has a central database and uses short messaging services (SMS) to send search terms that must have an exact match. Social media has been another approach capable of mobilizing volunteers to spread the information concerning the missing child at a fast rate. However, it lacks credibility since any one can author the information. A “CodeSearch” application was introduced in Canada which uses global positioning system (GPS) to send geo-targeted alerts to its subscribers. However, most people tend to have their GPS turned off unless when in use. It is also limited to employees of the CodeSearch partners. This research aims at introducing an expert system that uses ID3 algorithm to populate its knowledge base and an interactive search using the same algorithm. This allows users to interactively search the database, enter details about their missing loved ones if not yet found and notify them whenever the person is found. The search is based on the person’s phenotypes as they cut across the human race. The research is a form of applied research. The sample size was computed through convenience non-probability sampling. Most of the respondents recommended a proper system hence the reason for creating this prototype. The prototype is developed using V-Process methodology since the clarity of the user requirements was high and the technical expertise needed was readily available. The prototype produced 99% accuracy in tracing the missing children in the sample used.
  • Item
    Real – time sentiment analysis for detection of terrorist activities in Kenya
    (Strathmore University, 2016) Ngoge, Lucas Achuku
    Terrorism has become a subject of concern to many people in Kenya today. Majority of people are worried lot because they don’t know when they will become victims of terrorists’ activities. Corruption, porous border and luck of government in the neighboring Somali, have made Kenya a potential target for terrorists’. The advancement in technology has brought a new era in terrorism where Online Social Networks such as Twitter, Facebook has driven the increase use of the internet by terrorist organizations and their supporters for a wide range of purposes including recruitment, financing, propaganda, incitement to commit acts of terrorism and the gathering and dissemination of information for terrorist activities. Although the Kenya government improved its ability to fight terrorism but the changing pattern of terrorist activities, human errors and delayed crime analyses have given criminals more time to destroy evidence and escape arrest. The evolution of computerized systems has made tracking of terrorist’ activities easier. This has helped the law enforcement officers to speed up the process of solving crimes. In this research data was collected from twitter then followed by sentiment analysis on tweets collected to derive rules for the real-time classifier. Geographic analysis was done to reveal a correlation between the tweets and the terrorist’ activities as portrayed by the map. The main objective of this research is to develop a model that will be used to establish crime patterns associated with terrorist activities using sentiment information deduced from twitter data. To achieve this objective, 346 tweets related to terrorism were collected, cleansed and stored in a database for a period of 7 days. This data was then used as features for training and development of the model which will then be used to carry out real time sentiment analysis on twitter data. The model was tested and it was able to classify text correctly into positive, negative and neutral classes with an accuracy score of 73%.