Considerations of an efficiency-intelligent geo-localised mobile application for personalised SME market predictions

Abstract

The few applications used by Small and Medium Scale Enterprises (SMEs’) businesses lack efficiency and the appropriate intelligence to save them from price instability, inventory carrying costs, excess inventory, wrong decision making, inaccurate monitoring of stock levels, etc. through predictive analytics. The study explored various Artificial Intelligence Machine Learning (AI/ML) models and data structure array types that could be used with the day-to-day local weather conditions of low and high temperatures to predict market parameters and aid SMEs with predictive data to use for combating wrong decision-making, inaccurate business monitoring and excess inventory, etc. Among the ML models explored included sequential minimal optimisation, iterative reweighted least-squares, Fan-Chen-Lin support vector regression, linear regression newton method and multivariate linear regression Ordinary least squares for a multivariate linear regression and logistic regression. The models were compiled using visual C# and Accord.Net libraries. Multivariate linear regression Ordinary least squares models recorded the least predictive accuracy loss, for the test quantity prediction test samples, and varying acceptable square loss values, for usage in geo-localised mobile intelligent systems for SME predictions due to their favourable scores. The jagged array overall performed better than the multi-dimensional array on some time and space complexity tests. This work is contributing to the body of knowledge by evaluatively suggesting better data structures and ML models for building intelligent systems in Xamarin forms using C# and small data for the model training for applications in mobile phone systems that will aid SMEs’ in adjusting spending and sales targets.