ZS-data-science-challenge

3 solution files, 1st solution LB-0.847, 2nd solution LB-0.921, 3rd solution (using stacking) LB-0.956. Although 2nd and 3rd solution i created after the competition got over hope you can find something useful through this. zsstack.py contains the code for solution , I use target mean encoding to encode my categorical features , then produce a soltion using stacking. the base models i used are: MLP based classifier,catboost,adaboost,random forest,Lightgbm and XGboost. At 2nd level i.e. i tried various models as meta learners the best results were given by MLP based classifier with an accuarcay of 78% and LB score of 0.956 the LB rank at this score was 9th all india although i was not able to submit this solution as i realised later that my handling of missing values was not proper plus targget mean encoding really helped me in improving my results significantly You can try other modifications i neglected the adoption of time series analysis and i personally feel using time series will definetly help you break past 0.97 mark.Good Luck .Have fun and don't forget to give credit if you happen to use this code.Datazs.csv is the original dataset provided in the competition.