US20260236837
2026-08-13
Physics
G06N20/00
The patent application describes a system and method for creating machine learning-driven regressive forecasting models optimized for user-declared lightweight datasets. This innovation allows users to select features and targets, then trains a pool of AI/ML models with relevant data. The system identifies the best model using predetermined evaluation criteria and generates forecast results for the selected targets. This approach is particularly beneficial for environments where data is limited or volatile, such as in financial enterprises.
Traditional forecasting methods often struggle with the limited and volatile nature of data available to financial enterprises, leading to inaccuracies in predicting operations like monthly volumes. Large datasets are typically required for reliable forecasts, but smaller datasets can lead to errors. Additionally, the complexity of advanced machine learning techniques usually demands extensive coding knowledge, limiting accessibility for many potential users. Consequently, there is a need for a more accessible tool that simplifies machine learning tasks without requiring deep technical expertise.
The disclosed systems and methods enable users to build, optimize, and deploy forecasting models through automation, eliminating the need for extensive coding skills. An intuitive interface allows users to upload data, select features, and define targets with ease. The system automates processes like data preparation, feature engineering, and model optimization, making advanced machine learning accessible even to those with limited technical knowledge. This approach ensures that users can develop high-performance forecasting models quickly and efficiently.
Upon receiving a dataset from a user, the system presents a feature set and target prediction set through a user interface. Users select features and targets, and the system trains various AI/ML models with the selected data. The best-performing model is then chosen based on evaluation criteria, and forecast results are generated for each target. The system's operations are facilitated by a non-transitory machine-readable medium containing executable instructions for these processes.
The system includes components for data collection, preprocessing, and feature engineering. It is designed to work with lightweight datasets, such as those with monthly or daily intervals. The system automates intricate processes like feature engineering and model selection, reducing error potential and cognitive load. This allows non-experts to achieve accurate results without deep technical knowledge. The system's flexibility supports various applications, including sales forecasting, resource allocation, and government sector analyses.