Data Fusion Practical Work

Practical work for data fusion subject

This Practical Work was made with the objective create a dataset, and upon that dataset make an study to train a classifier capable of detect the type of application that a mobile phone is using based on some of the information and sensors avaliable on the device.

For this project a Dataset was collected using the build in features of HomeAssitant to collect mobile phones information and then creating a dataset from that data. A total of 51 features where extracted from each mobile phone and the a total of six mobile phones from different brands where used to collect the data.

February 2023 Barcelos, Portugal

Professor of Fusion Data (Vitor Oliveira)

Topic:

Study and manipulation of data from different sources for class classification using artificial intelligence (AI) algorithms. The data acquisition can be done using datasets available on the web or else acquired through systems with sensors. The dataset must have at least 5 features (columns with different characteristics).

Project Requirements

(1) Analysis of Dataset Features.
(2) Preparation of the data to put into an AI model.
(3) Study of the most suitable Ai model for the chosen dataset.
(4) Application of the selected Ai model to the dataset with and without the application of PCA/LDA techniques.
(5) Analysis of the Ai model training (Confusion matrix, learning curves, etc...).
(6) Study of the most important features of the dataset (feature importance) to improve the accuracy of the Ai model.
(7) Feature Importance self Obtained. (By own algorithms).

Valorization of the Process

(8) Hardware for acquiring the dataset.
(9) Application of data combination techniques (example: data alignment, downsampling/upsampling, etc...).
(10) Application of data quality improvement techniques (example: filters).
(11) Other approaches...

Characteristics

● Code in Python.
● Identification and explanation of the options taken, framing them in the various topics taught in class.
● Result of the classifiers trained and comments, in each step of the algorithm (graphical representations should be given preference).

Delivery:

• Jupyter notebook
• Dataset for testing the developed algorithm