9/27/21 1 Boston University –CS767, Machine Learning, F. Alizadeh-Shabdiz Classification Assessment Data Science with Python CS677 Farshid Alizadeh-Shabdiz, PhD, MBA Alizadeh@bu.edu Fall 2021 1 1 Boston University –CS767, Machine Learning, F. Alizadeh-Shabdiz Classification Assessment 2 ©2013 Skyhook Wireless Inc. Boston University –CS677, Data Science with Python, F. Alizadeh-Shabdiz3 Let us assume the output of logistic regression is 0.37 and 0.84. If all the samples are decided correctly, it is perfect! What if the result is not perfect! How to Quantify the Fitness in Logistic Regression? 3 ©2013 Skyhook Wireless Inc. Boston University –CS677, Data Science with Python, F. Alizadeh-Shabdiz4 Reference: Wikipedia 4 ©2013 Skyhook Wireless Inc. Boston University –CS677, Data Science with Python, F. Alizadeh-Shabdiz5 5 ©2013 Skyhook Wireless Inc. Boston University –CS677, Data Science with Python, F. Alizadeh-Shabdiz6 6 9/27/21 2 ©2013 Skyhook Wireless Inc. Boston University –CS677, Data Science with Python, F. Alizadeh-Shabdiz7 7 ©2013 Skyhook Wireless Inc. Boston University –CS677, Data Science with Python, F. Alizadeh-Shabdiz8 8 ©2013 Skyhook Wireless Inc. Boston University –CS677, Data Science with Python, F. Alizadeh-Shabdiz9 9 ©2013 Skyhook Wireless Inc. Boston University –CS677, Data Science with Python, F. Alizadeh-Shabdiz10 10 ©2013 Skyhook Wireless Inc. Boston University –CS677, Data Science with Python, F. Alizadeh-Shabdiz11 11 ©2013 Skyhook Wireless Inc. Boston University –CS677, Data Science with Python, F. Alizadeh-Shabdiz12 12 9/27/21 3 ©2013 Skyhook Wireless Inc. Boston University –CS677, Data Science with Python, F. Alizadeh-Shabdiz13 13 ©2013 Skyhook Wireless Inc. Boston University –CS677, Data Science with Python, F. Alizadeh-Shabdiz14 14 ©2013 Skyhook Wireless Inc. Boston University –CS677, Data Science with Python, F. Alizadeh-Shabdiz15Boston University –CS767, Machine Learning, F. Alizadeh-Shabdiz 15 ©2013 Skyhook Wireless Inc. Boston University –CS677, Data Science with Python, F. Alizadeh-Shabdiz16Boston University –CS767, Machine Learning, F. Alizadeh-Shabdiz 16 ©2013 Skyhook Wireless Inc. Boston University –CS677, Data Science with Python, F. Alizadeh-Shabdiz17Boston University –CS767, Machine Learning, F. Alizadeh-Shabdiz 17 ©2013 Skyhook Wireless Inc. Boston University –CS677, Data Science with Python, F. Alizadeh-Shabdiz18 Cross entropy is an appropriate tool to compare classifiers But Cross entropy doesn’t relate to a meaningful fact In the absolute sense doesn’t provide any insight into classifier It is not bounded. It goes from zero to infinity Cross Entropy as an Assessment Tool 18