This study sought to understand how machine learning could facilitate the tracking of bees and address the lack of automated bee-tracking tools that can count bee behaviour and movement from positional information. Harmonic radar tracking datasets of bumblebee movement were used to predict flight tasks. Random forest (RF), Support Vector Machine (SVM) and neural network (NN) algorithms were trained on the dataset and their performance was evaluated. Unsupervised clustering (lacking any human labels) was performed to investigate whether a simple binary classification of bee tasks (foraging or exploring) could be replaced with a multiclassification model with more complex behaviour modelling. Comparisons of optical and thermal camera systems were also undertaken and found that thermal cameras, which are less affected by sub-optimal lighting, are more suitable for automated bee flight characterisation, (inwards, outwards, and hovering) at the entrance to the hive. Gaussian mixture models and a Kalman filter were used to extract bee flights from recordings from bee hives. The recorded flights were pre-processed into a dataset to train SVM, RF, and NN algorithms to predict the three flight types. Finally, a Doppler radar was used to record bee entry and exit activity from a hive. The data was processed using Log Area Ratios, derived from Linear Prediction Coefficients, to create a dataset for training machine models. The goal was to create a system that could function in real-time using a Raspberry Pi to classify the activity at the entrance of a hive. This study demonstrates that machine learning could automate a data-intensive field of study and provide meaningful insights into the activities of bee species with uses in the apicultural sector, including research and conservation.