model-performance 0.0.2

Model Analysis and Performance Package

# Model Performance

# Introduction

This package calculates model performance metrics and confusion metrics at decile level of prediciton. Instead of calculating the model performance at default probability cut-off of 0.5, this function divides prediction in 10 equal buckets and for each bucket provides the model performance.

# Input

The function “conf_matrix” requires an input dataset name, the prediction feature name and the target feature name

# Results

decile_pred - Deciles of predictions. cnt - Count of observations in each decile. max_pred - Maximum probablity in each decile. resp - Actual sum of target in each decile. non_resp - Actual sum of non-target in each decile. cumresp - Actual cumulative sum of target in each decile. cumcnt - Cumulative count of observations in each decile. resp_rate - Target rate of each decile. lift - Lift in target rate of each decile. ex - By targeting top 20% customers we achieve 4x lift than random targeting. tp - True positive at each decile cut-off. fp - False positive at each decile cut-off. tn - True negative at each decile cut-off. fn - False negative at each decile cut-off. precision - Precision at each decile cut-off. recall - Recall at each decile cut-off. fscore - Fscore at each decile cut-off. ks - Maximum separation between target and non-target at each decile.

# Contact Mrinal Shankar (https://github.com/mrinal-shankar)