The work is conducted in collaboration with Toyota Material Handling in Mjölby andfocuses on developing a scalable, end-to-end data pipeline capable of transforming rawmulti-source sensor data into a structured format suitable for machine-learning-based energyestimation. Despite the availability of large heterogeneous data for electric forklifts,there is a lack of efficient methods to handle such data. This thesis investigates methods forhandling heterogeneous time-series data with varying update frequencies in the context ofpredicting the energy consumption of electric forklifts.The primary contribution of the thesis is a data-processing framework that aligns andintegrates time-series signals — such as current, voltage, and vehicle speed — originating from different sampling rates and acquisition systems. A segmentation and phase identificationmethodology is proposed to isolate operational states within the travel scenario,enabling targeted feature extraction relevant to energy-use prediction. Machinelearning techniques are applied at multiple stages of the workflow, including data fusion,phase classification, and supervised estimation of energy consumption. The transformedand derived results are stored and visualized using structured Excel outputs to supportfurther analysis and validation.The outcomes demonstrate that consistent preprocessing of heterogeneous sensorstreams is critical for accurate energy-consumption modeling, and provide a foundationfor future predictive systems in material-handling applications.