The dataset

Upon releasing their findings to the scientific community in 2012, researchers later made the data public from the LHC experiments where they observed - and identified - a signal which is indicative of the Higgs-Boson particle. However, amidst the positive findings is a lot of background noise which causes an imbalance within the dataset. Our task as data scientist is to build a machine learning model which can accurately identify the Higgs-Boson particle from background noise. Already, you should be thinking about how this question is phrased which would be indicative of binary classification (that is, is this example the Higgs-Boson versus background noise?).

This file is 2.6 gigs (uncompressed) and contains 11 million examples that have been labeled as 0 - background noise and 1 - Higgs-Boson. First, you will need to uncompress this file and then we will begin loading the data into Spark for processing and analysis. There are 29 total fields which make up the dataset:

• Field 1: Class label (1 = signal for Higgs-Boson, 2 = background noise)
• Fields 2-22: 21 "low-level" features that come from the collision detectors
• Fields 23-29: seven "high-level" features that have been hand-derived by particle physicists to help classify the particle into its appropriate class (Higgs or background noise)

Later in this chapter, we cover a Deep Neural Network (DNN) example that will attempt to learn these hand-derived features through layers of non-linear transformations to the input data.

Note that for the purposes of this chapter, we will work with a subset of the data, the first 100,000 rows, but all the code we show would also work on the original dataset.