What are the key points?

sktime offers a unified Python API for time series forecasting, classification, and regression tasks. The library features pipeline workflows for data imputation, detrending, and deseasonalizing HVAC sensor readings. ExponentialSmoothing achieved 0.584°C MAE, while cross-validation using temporal splitters resulted in 0.606°C MAE.

Building Time-Series Models with sktime

•sktime offers a unified Python API for time series forecasting, classification, and regression tasks.
•The library features pipeline workflows for data imputation, detrending, and deseasonalizing HVAC sensor readings.
•ExponentialSmoothing achieved 0.584°C MAE, while cross-validation using temporal splitters resulted in 0.606°C MAE.

The sktime library provides a Python framework for time series analysis using a scikit-learn-style API. It handles series, panel, and hierarchical data structures through consistent interfaces for forecasting, classification, and regression.

A forecasting workflow using 90 days of hourly HVAC sensor data utilizes temporal_train_test_split() to ensure chronological data separation and avoid leakage. The TransformedTargetForecaster pipeline integrates Imputer, Detrender, and Deseasonalizer components before applying models like ExponentialSmoothing or ARIMA.

Testing results on temperature forecasting show an MAE of 0.584°C and MAPE of 2.40% for ExponentialSmoothing, while ARIMA yielded 0.586°C. Cross-validation via ExpandingWindowSplitter and SlidingWindowSplitter produced an MAE of 0.606°C.

The sktime library provides a Python framework for time series analysis using a scikit-learn-style API. It handles series, panel, and hierarchical data structures through consistent interfaces for forecasting, classification, and regression.

A forecasting workflow using 90 days of hourly HVAC sensor data utilizes temporal_train_test_split() to ensure chronological data separation and avoid leakage. The TransformedTargetForecaster pipeline integrates Imputer, Detrender, and Deseasonalizer components before applying models like ExponentialSmoothing or ARIMA.

Testing results on temperature forecasting show an MAE of 0.584°C and MAPE of 2.40% for ExponentialSmoothing, while ARIMA yielded 0.586°C. Cross-validation via ExpandingWindowSplitter and SlidingWindowSplitter produced an MAE of 0.606°C.