This is an exploratory machine learning project aimed to compare the performance of different Neural Networks in classifying liquid-immersed capacitive sensor electrode signals. The authors' aim is to systematically compare multiple different approaches to the same task. Three primary neural networks will be trained: Multi-Layer Perceptrons (MLP), Convolutional Neural Networks (CNN) and Recurrent Neural Networks (RNN). Metrics such as accuracy, precision, recall and F1-score, will be used to determine the best neural network model for this task.
The dataset is retrieved from Kaggle, but was originally derived from Mahdi Saleh, Imad H. Elhajj and Daniel Asmar’s study on digital sensor signals.
The dataset is composed of digital signals obtained from a capacitive sensor electrodes that are immersed in water or in oil. Each signal, stored in one row, is composed of 10 consecutive intensity values and a label in the last column. The label is +1 for a water-immersed sensor electrode and -1 for an oil-immersed sensor electrode. This dataset should be used to train a classifier to infer the type of material in which an electrode is immersed in (water or oil), given a sample signal composed of 10 consecutive values.
The final models produced from the project are uploaded on the authors' Github repository under the folder application_files as .h5 files. There are a total of three models: 1 for MLP, 1 for CNN, and 1 for RNN. The models have the filenames as follows:
- final_mlp.h5
- final_cnn.h5
- final_rnn.h5
For the purpose of running the separate notebook for application, these files do not need to be downloaded.