On the Development of Machine Learning – Based Application Framework for Enhancing Performance of Livestock Mobile Application Systems in Poor Internet Service Areas

Abstract

In this paper, authors developed an intelligent subsystem which manages training set, finds high accuracy models, selects best model to be used, computes prediction, stores in the database, and sends to the user interface through internet during online mode, and in offline mode through developed log file and filtering method. The intelligent subsystem is one of solutions which support mobile phone systems to be executed offline, on mobile device. Prediction results can be locally stored in the database and log file while in presence of a fairly good connection environment. Thereafter, offline predictions are made available when a poor quality in connection comes. System development covers intelligent subsystem, MySQL database development, log file, filtering of information, and Android application. Apart from viewing predictions basing on ElasticNet algorithm, the system allows a user to register, login, and access livestock market, information portal, information request, information responses, and submit daily records.The filtering techniques are used to select part of information from the log file. The log file created on the last online activity is used to serve all the offline operations as follows: -1) Once the user has selected the offline option in the app's interface, he/she is directed to select the breed, sex and grade. 2) Upon submission of his/her input i.e. breed, sex and grade, the system reads the log and through the filtering algorithm, the predicted price is captured and displayed to the user.Overall, this paper makes the following six key contributions; (1) Creating a mechanism to select a subset of information between livestock keeper and server. The subset of information is used when internet fails. (2) To design and implement machine learning based sub-system that is able to select frequently asked requests and make them available to livestock keepers during the offline mode. (3) To develop a subsystem, this is ad hoc and performs model selection. (4) Model creation that evaluates more than one algorithm in our setting or context. Others compared algorithms in their settings. (5) Introduction of a smaller database that replicates information, which smallholder farmers requests online and stores them to be available during offline state. (6) Using database of the phone to store vital information.