mozilladatascience / outreachy-datascience-2018 Goto Github PK

A better understanding is needed on how the attributes relate to device failure.

• How do the attributes relate to device failure?
• Is there a single attribute, or a combination of the attributes, that follows device failure?

It is important to determine if there are seasonality effects regarding the telemetry data.

• What seasonality effects are observed in the data?
• Is there seasonality influences in device failure?

The relationship between the attributes is poorly understood. A better understanding is needed in this regard for the data we are collecting.

• What is the correlation between the attributes?
• How do the scales differ or are similar between attributes?

A big concern is the overall integrity of the telemetry data, which needs to be addressed through investigation.

• Is there missing data?
• Is the data consistent?
• Are there bugs in the data?

The desire is to try to utilize the telemetry attributes to predict device failure in the field. This requires generating a feature set for use in modeling.

• What are good features for modeling device failure?
• How would you choose the best features?

The administration wants to better understand the social dynamic of the high school. The dataset contains a relevant column called friends, which is the number of friends of the student.

• What is the distribution of the number of friends?
• What are the gender breakdown of friends?
• What is the breakdown by race?

The current telemetry payload is too large. We need a good representation that approximates the dataset, with a reduced number of columns.

• What are some methods for reducing the size of the dataset?
• What is the trade-off between dataset size and fidelity to original dataset

The current dataset stores parent information in the parents field. One value is my mother and my father. It is desirable to break out this information into single caregivers. e.g., my mother and my father becomes mother and father, whereas my mother only is just mother.

• What is the distribution of each single caregiver type? e.g., What is the distribution of mother caregiver?

A big concern is the overall integrity of the school dataset, which needs to be addressed through investigation.

• Is there missing data?
• Is the data consistent?
• Are there bugs in the data?

A better understanding is needed regarding the behavior of failed devices.

• How many devices failed?
• How often do they fail?
• When do they fail?