Weather-Driven Solar Intensity Forecasting for Smart Homes : A Comparative Machine Learning Study

Solar intensity forecasting can help residential photovoltaic systems anticipate solar availability and plan electricity use. This paper presents a detailed account of the approach described in the supplied seminar report, which predicts global horizontal irradiance (GHI) using weather-related variables. The reported pipeline includes data cleaning, normalization, correlation analysis, an 80:20 training-test split, and a comparison of conventional machine-learning algorithms with a fully connected neural network. The considered models are Linear Regression, Quadratic Regression, Naive Bayes, k-Nearest Neighbours (KNN), Support Vector Machine (SVM), Gradient Boosting Machine (GBM), and a neural network. The report records the lowest mean squared error (MSE) for the neural network at 1,675.94, followed by KNN at 2,587.07. These results indicate that the neural network performed best among the evaluated alternatives in the reported experiment. Some dataset and training details are not specified sufficiently for exact reproduction; therefore, the paper presents the reported findings without claiming independent implementation or validation.