Implementation to predict human movement towards a specific Access Point

This is the implementation for my bachelor's thesis "Machine Learning-based User Movement Prediciton in Layer 2 Networks" at Hasso-Plattner-Institute.

The goal of this thesis is to find out, if human movement towards a specific Access Point can be predicted.

Getting the data

1. To download the files needed for the analysis and machine learning part, please go to dataset.
2. Unzip the .zip file into the root folder of this repository.

Working with the data

1. Make sure you have downloaded the data as described in the previous step.
2. Execute any of the following scripts:
   • data-ana.py: To get an overview of the folder structure
   • aggregate_file.py: To aggregate all files of a floor to one file.
   • count_wifi_lines.py: To get to know, how many Wi-Fi data there is for the floor with the most files/traces (it uses the aggregate file).

Preparation

Execute preparation.ipynb to prepare the data of the floor with the most traces for the LSTM model.

Tune, train and test model

Execute lstm.ipynb for tuning, training and testing the LSTM model.

Evaluate the model

Execute evaluation.ipynb to evaluate the model and compare it to a heuristic approach. The results can be seen in comparison_ml_heuristic_1_to_5.pdf, comparison_ml_heuristic_3.pdf and heuristic_plot.pdf.