A Functional Data-Driven Approach for Accurate Employee Presence Forecasting

Abstract

In smart workplaces, figuring out when employees will be there is important for making the best use of room and resources. This research shows a smart way to make predictions using a dataset of employee presences that includes information about time and place. After data preprocessing gets rid of null values and duplicates, exploratory analysis and association visualization are used to see patterns in how people behave. Several prediction methods are used, such as AdaBoost, SARIMAX, FTSA, LSTM, DFMM, RobFTS, CNN, and an ensemble Voting Regressor that combines XGBoost, Gradient Boosting, and K-Nearest Neighbors. Functional Data Analysis is used to describe how things depend on time and how occupancy patterns change over time. The DFMM model gets 4.68% MAPE and a R² value of 91.3%, which means it is very accurate. On the other hand, the Voting Regressor gets better generalization with a R² value of 93.1%. Explainable AI methods, like LIME and SHAP, are used to figure out what forecasts mean and find features that are important. For real-world use, the system is built on the Flask framework, which provides a simple web interface with safe user signup and signin through SQLite, file-based user input, processing on the back end, and interactive visualization. Users upload test files and choose the time frames for the predictions. The system then shows weekly and daily employee presence forecasts, including predictions for the next four or eight weeks. This lets managers make smart choices about how to plan the workplace and use resources most efficiently