This is a mini-project on feature engineering and machine learning.
Introduction on Task and Context:
Different from traditional top-down performance evaluation where mangers evaluate their subordinates, 360-degree performance appraisal asks everyone in the team to evaluate everyone else. This process generates very rich information about the team. I use individual characteristics and features extracted from the 360-degree performance appraisal to predict promotion outcomes.
Datasets:
dataset.csv contains individual characteristics and promotion outcomes.
ratings.csv contains ratings from 360-degree performance appraisal.
These two datasets will be uploaded after careful data desensitization to protect privacy.
iPython Notebooks:
(1) Machine Learning.ipynb
Using Scikit-learn, I predict promotion outcomes employing logistic regression, decision tree model, and random forest. GridSearchCV and RandomizedSearchCV are used to tune hyperparameters in these models.
In addition, I use the feature importance measure to evaluate the usefulness of features extracted from the from 360-degree performance appraisal.
(2) Feature Engineering.ipynb
Essentially, every individual in the team is a human sensor. In the first level, information generated by the 360-degree appraisal tells us about individual performances. In the second level, it also informs about the sensors themselves: how people in the team perceive and evaluate their teammates. For example, to which extent one's evaluation deviates from the consensus. In this note, I extract features in the second level. For an individual in the team, I measure the degree to which his/her evaluations are different from the team's consensus. The team's consensus is measured by averaging everyone's evaluations. In addition, I also measure the extent to which a team member's evaluation differs from the team leader's.
I explore predictive power of these new features. In particular, to which extent these features can help in predicting promotion outcomes in the team.