Design and implementation of a smart weather application with weather prediction and data analysis : bachelor's thesis on smart data driven weather systems

Abstract

Most weather applications show too much technical data and do not help users to understand what the weather means for their daily life decisions. This thesis investigates whether simple statistical models in a lightweight web interface can bridge the gap between raw weatherdata and practical daily guidance, by building a simple and easy to use web application that shows current weather information and predicts the next day's temperature.The application is built using Python and Flask for the backend and HTML, CSS, and JavaScript for the frontend. Weather data are collected from the OpenWeather API. Three prediction methods are used and compared: Simple Moving Average, Weighted Moving Average, and Simple Linear Regression.

The application also gives plain language recommendations such as what to wear or whether to carry an umbrella based on the currentweather conditions.The application is tested on many cities around the world. Linear Regression achieved the lowest Mean Absolute Error and therefore provided the most accurate temperature predictions among the three implemented models. When tested using API forecast data, ithad the lowest error (0.34°C) and was 43 percent more accurate than the Simple MovingAverage model. In the real-world London test, which used actual next-day temperatures, Linear Regression also gave the best results, with an average error of 2.16°C. This shows that it was the most accurate model in both tests. All twelve functional requirements are successfully met. The results show that a simple and lightweight weather application usingbasic statistical methods can provide clear, useful, and easy to understand weather information to everyday users.