Skip to content
/ thesis-pub Public

Master's thesis on the topic “Application of decision tree models to solve real problems”

0 stars 0 forks Branches Tags Activity
Star
Notifications You must be signed in to change notification settings

sukhrobyangibaev/thesis-pub

BranchesTags

Repository files navigation

Applying Decision tree models to solve real-life problems

Understanding Decision Tree Models: Theory and Methodology.

Implementing a Universal Decision Tree Framework and Experimental Validation.

Data CART C4.5 Extra Trees Classifier Gradient Boosting Hist Gradient Boosting Random Forest Adaboost
Annual Profit 81.6 81.3 85.5 87.3 87.4 86.4 86.9
Breast Cancer 93.4 92.7 96.4 95.6 94.9 94.9 95.6
Credit Approval 87.8 78.6 87.8 91.6 89.3 90.0 90.8
Energy Efficiency 98.7 98.0 99.4 98.7 98.7 99.4 84.4
Glass Identification 69.8 72.1 79.1 72.1 72.1 76.7 67.4
Ionosphere 87.1 90.0 95.7 95.7 94.3 95.7 97.1
Iris Plants 96.7 96.7 96.7 100.0 96.7 96.7 96.7
Nursery 99.5 99.4 97.8 100.0 98.5 98.4 87.9
Spam 90.4 92.2 95.7 95.7 95.2 95.5 94.8
Zoo 95.0 95.0 100.0 95.0 100.0 100.0 95.0
90 89.6 93.4 85.6 92.7 93.3 89.6

Exploring Open Dota API and Steam API for Data Acquisition.

Implementing Data Miners for Real-Time Data Acquisition

Miners For Public Matches

RealtimeStats.py

This Python script continuously fetches real-time statistics for top live Dota 2 games from the Dota 2 API and stores them in a MongoDB collection for further analysis.

GetWinImportToDB.py

This Python script retrieves match details for unprocessed matches stored in a MongoDB collection and updates the collection with the winning team information.

Miners For League Matches

LiveLeagueGames.py

This script continuously fetches live league games data from the Dota 2 API and stores them in a MongoDB collection for real-time analysis.

GetLeagueWin.py

This Python script retrieves match details from the Dota 2 API for matches stored in a MongoDB collection. It iterates through unprocessed match IDs, retrieves match details, and updates the database with the winning team information.

Training Models (Public Matches)

This section presents the performance of various decision tree-based models trained on different sizes of public match datasets.

Notebooks

Selected Features: 313 timestamp,game_time,score,net_worth,team_2_player_0_heroid,team_2_player_0_level,team_2_player_0_kill_count,team_2_player_0_death_count,team_2_player_0_assists_count,team_2_player_0_denies_count,team_2_player_0_lh_count,team_2_player_0_gold,team_2_player_0_net_worth,team_2_player_0_item_0,team_2_player_0_item_1,team_2_player_0_item_2,team_2_player_0_item_3,team_2_player_0_item_4,team_2_player_0_item_5,team_2_player_0_item_6,team_2_player_0_item_7,team_2_player_0_item_8,team_2_player_1_heroid,team_2_player_1_level,team_2_player_1_kill_count,team_2_player_1_death_count,team_2_player_1_assists_count,team_2_player_1_denies_count,team_2_player_1_lh_count,team_2_player_1_gold,team_2_player_1_net_worth,team_2_player_1_item_0,team_2_player_1_item_1,team_2_player_1_item_2,team_2_player_1_item_3,team_2_player_1_item_4,team_2_player_1_item_5,team_2_player_1_item_6,team_2_player_1_item_7,team_2_player_1_item_8,team_2_player_2_heroid,team_2_player_2_level,team_2_player_2_kill_count,team_2_player_2_death_count,team_2_player_2_assists_count,team_2_player_2_denies_count,team_2_player_2_lh_count,team_2_player_2_gold,team_2_player_2_net_worth,team_2_player_2_item_0,team_2_player_2_item_1,team_2_player_2_item_2,team_2_player_2_item_3,team_2_player_2_item_4,team_2_player_2_item_5,team_2_player_2_item_6,team_2_player_2_item_7,team_2_player_2_item_8,team_2_player_3_heroid,team_2_player_3_level,team_2_player_3_kill_count,team_2_player_3_death_count,team_2_player_3_assists_count,team_2_player_3_denies_count,team_2_player_3_lh_count,team_2_player_3_gold,team_2_player_3_net_worth,team_2_player_3_item_0,team_2_player_3_item_1,team_2_player_3_item_2,team_2_player_3_item_3,team_2_player_3_item_4,team_2_player_3_item_5,team_2_player_3_item_6,team_2_player_3_item_7,team_2_player_3_item_8,team_2_player_4_heroid,team_2_player_4_level,team_2_player_4_kill_count,team_2_player_4_death_count,team_2_player_4_assists_count,team_2_player_4_denies_count,team_2_player_4_lh_count,team_2_player_4_gold,team_2_player_4_net_worth,team_2_player_4_item_0,team_2_player_4_item_1,team_2_player_4_item_2,team_2_player_4_item_3,team_2_player_4_item_4,team_2_player_4_item_5,team_2_player_4_item_6,team_2_player_4_item_7,team_2_player_4_item_8,team_3_player_0_heroid,team_3_player_0_level,team_3_player_0_kill_count,team_3_player_0_death_count,team_3_player_0_assists_count,team_3_player_0_denies_count,team_3_player_0_lh_count,team_3_player_0_gold,team_3_player_0_net_worth,team_3_player_0_item_0,team_3_player_0_item_1,team_3_player_0_item_2,team_3_player_0_item_3,team_3_player_0_item_4,team_3_player_0_item_5,team_3_player_0_item_6,team_3_player_0_item_7,team_3_player_0_item_8,team_3_player_1_heroid,team_3_player_1_level,team_3_player_1_kill_count,team_3_player_1_death_count,team_3_player_1_assists_count,team_3_player_1_denies_count,team_3_player_1_lh_count,team_3_player_1_gold,team_3_player_1_net_worth,team_3_player_1_item_0,team_3_player_1_item_1,team_3_player_1_item_2,team_3_player_1_item_3,team_3_player_1_item_4,team_3_player_1_item_5,team_3_player_1_item_6,team_3_player_1_item_7,team_3_player_1_item_8,team_3_player_2_heroid,team_3_player_2_level,team_3_player_2_kill_count,team_3_player_2_death_count,team_3_player_2_assists_count,team_3_player_2_denies_count,team_3_player_2_lh_count,team_3_player_2_gold,team_3_player_2_net_worth,team_3_player_2_item_0,team_3_player_2_item_1,team_3_player_2_item_2,team_3_player_2_item_3,team_3_player_2_item_4,team_3_player_2_item_5,team_3_player_2_item_6,team_3_player_2_item_7,team_3_player_2_item_8,team_3_player_3_heroid,team_3_player_3_level,team_3_player_3_kill_count,team_3_player_3_death_count,team_3_player_3_assists_count,team_3_player_3_denies_count,team_3_player_3_lh_count,team_3_player_3_gold,team_3_player_3_net_worth,team_3_player_3_item_0,team_3_player_3_item_1,team_3_player_3_item_2,team_3_player_3_item_3,team_3_player_3_item_4,team_3_player_3_item_5,team_3_player_3_item_6,team_3_player_3_item_7,team_3_player_3_item_8,team_3_player_4_heroid,team_3_player_4_level,team_3_player_4_kill_count,team_3_player_4_death_count,team_3_player_4_assists_count,team_3_player_4_denies_count,team_3_player_4_lh_count,team_3_player_4_gold,team_3_player_4_net_worth,team_3_player_4_item_0,team_3_player_4_item_1,team_3_player_4_item_2,team_3_player_4_item_3,team_3_player_4_item_4,team_3_player_4_item_5,team_3_player_4_item_6,team_3_player_4_item_7,team_3_player_4_item_8,graph_gold_0,graph_gold_1,graph_gold_2,graph_gold_3,graph_gold_4,graph_gold_5,graph_gold_6,graph_gold_7,graph_gold_8,graph_gold_9,graph_gold_10,graph_gold_11,graph_gold_12,graph_gold_13,graph_gold_14,graph_gold_15,graph_gold_16,graph_gold_17,graph_gold_18,graph_gold_19,graph_gold_20,graph_gold_21,graph_gold_22,graph_gold_23,graph_gold_24,graph_gold_25,graph_gold_26,graph_gold_27,graph_gold_28,graph_gold_29,graph_gold_30,graph_gold_31,graph_gold_32,graph_gold_33,graph_gold_34,graph_gold_35,graph_gold_36,graph_gold_37,graph_gold_38,graph_gold_39,graph_gold_40,graph_gold_41,graph_gold_42,graph_gold_43,graph_gold_44,graph_gold_45,graph_gold_46,graph_gold_47,graph_gold_48,graph_gold_49,graph_gold_50,graph_gold_51,graph_gold_52,graph_gold_53,graph_gold_54,graph_gold_55,graph_gold_56,graph_gold_57,graph_gold_58,graph_gold_59,graph_gold_60,graph_gold_61,graph_gold_62,graph_gold_63,graph_gold_64,graph_gold_65,graph_gold_66,graph_gold_67,graph_gold_68,graph_gold_69,graph_gold_70,graph_gold_71,graph_gold_72,graph_gold_73,graph_gold_74,graph_gold_75,graph_gold_76,graph_gold_77,graph_gold_78,graph_gold_79,graph_gold_80,graph_gold_81,graph_gold_82,graph_gold_83,graph_gold_84,graph_gold_85,graph_gold_86,graph_gold_87,graph_gold_88,graph_gold_89,graph_gold_90,graph_gold_91,graph_gold_92,graph_gold_93,graph_gold_94,graph_gold_95,graph_gold_96,graph_gold_97,graph_gold_98,graph_gold_99,graph_gold_100,graph_gold_101,graph_gold_102,graph_gold_103,graph_gold_104,graph_gold_105,graph_gold_106,graph_gold_107,graph_gold_108,graph_gold_109,graph_gold_110,graph_gold_111,graph_gold_112,graph_gold_113,graph_gold_114,graph_gold_115,graph_gold_116,graph_gold_117,graph_gold_118,graph_gold_119,graph_gold_120,graph_gold_121,graph_gold_122,graph_gold_123,graph_gold_124,graph_gold_125,graph_gold_126,graph_gold_127,winner

Model Performance

Data CART C4.5 Extra Trees Classifier Gradient Boosting Hist Gradient Boosting Random Forest Adaboost
~1k samples 65.6 61.4 66.6 64.5 71.8 71.8 66.6
~15k samples 89.7 89.5 93.0 93.6 98.7 91.3 72.3
~20k samples 77.6 76.5 85.8 83.1 89.1 83.2 70.6
~30k samples 75.8 75.8 85.2 81.2 87.9 81.1 69.3
~40k samples 75.1 74.7 85.4 82.1 88.0 81.7 70.6

Training Models (League Matches)

Selected Features: 158 duration radiant_series_wins dire_series_wins score tower_state barracks_state radiant_player_{i}_kills radiant_player_{i}_death radiant_player_{i}_assists radiant_player_{i}_last_hits radiant_player_{i}_gold radiant_player_{i}_level radiant_player_{i}_gold_per_min radiant_player_{i}_xp_per_min radiant_player_{i}_item0 radiant_player_{i}_item1 radiant_player_{i}_item2 radiant_player_{i}_item3 radiant_player_{i}_item4 radiant_player_{i}_item5 radiant_player_{i}_net_worth dire_player_{i}_death dire_player_{i}_kills dire_player_{i}_assists dire_player_{i}_last_hits dire_player_{i}_gold dire_player_{i}_level dire_player_{i}_gold_per_min dire_player_{i}_xp_per_min dire_player_{i}_item0 dire_player_{i}_item1 dire_player_{i}_item2 dire_player_{i}_item3 dire_player_{i}_item4 dire_player_{i}_item5 dire_player_{i}_net_worth net_worth winner

Model Performance

Data CART C4.5 Extra Trees Classifier Gradient Boosting Hist Gradient Boosting Random Forest Adaboost
~20k samples 84.1 83.6 96.4 89.6 94.6 94.6 71.6
~53k samples 79.6 79.9 96.0 81.6 86.7 92.7 67.8
~100k samples 81.9 81.9 95.7 76.4 81.6 93.3 61.2

Training and Evaluation on League Match Data with Match Duration Segmentation (<10min, 10-20min, >20min)

Selected Features: 186 duration,0_rts,1_rts,2_rts,3_rts,4_rts,5_rts,6_rts,7_rts,8_rts,9_rts,10_rts,0_dts,1_dts,2_dts,3_dts,4_dts,5_dts,6_dts,7_dts,8_dts,9_dts,10_dts,0_rbs,1_rbs,2_rbs,3_rbs,4_rbs,5_rbs,0_dbs,1_dbs,2_dbs,3_dbs,4_dbs,5_dbs,radiant_player_0_kills,radiant_player_0_death,radiant_player_0_assists,radiant_player_0_last_hits,radiant_player_0_gold,radiant_player_0_level,radiant_player_0_gold_per_min,radiant_player_0_xp_per_min,radiant_player_0_net_worth,radiant_player_0_item0,radiant_player_0_item1,radiant_player_0_item2,radiant_player_0_item3,radiant_player_0_item4,radiant_player_0_item5,radiant_player_1_kills,radiant_player_1_death,radiant_player_1_assists,radiant_player_1_last_hits,radiant_player_1_gold,radiant_player_1_level,radiant_player_1_gold_per_min,radiant_player_1_xp_per_min,radiant_player_1_net_worth,radiant_player_1_item0,radiant_player_1_item1,radiant_player_1_item2,radiant_player_1_item3,radiant_player_1_item4,radiant_player_1_item5,radiant_player_2_kills,radiant_player_2_death,radiant_player_2_assists,radiant_player_2_last_hits,radiant_player_2_gold,radiant_player_2_level,radiant_player_2_gold_per_min,radiant_player_2_xp_per_min,radiant_player_2_net_worth,radiant_player_2_item0,radiant_player_2_item1,radiant_player_2_item2,radiant_player_2_item3,radiant_player_2_item4,radiant_player_2_item5,radiant_player_3_kills,radiant_player_3_death,radiant_player_3_assists,radiant_player_3_last_hits,radiant_player_3_gold,radiant_player_3_level,radiant_player_3_gold_per_min,radiant_player_3_xp_per_min,radiant_player_3_net_worth,radiant_player_3_item0,radiant_player_3_item1,radiant_player_3_item2,radiant_player_3_item3,radiant_player_3_item4,radiant_player_3_item5,radiant_player_4_kills,radiant_player_4_death,radiant_player_4_assists,radiant_player_4_last_hits,radiant_player_4_gold,radiant_player_4_level,radiant_player_4_gold_per_min,radiant_player_4_xp_per_min,radiant_player_4_net_worth,radiant_player_4_item0,radiant_player_4_item1,radiant_player_4_item2,radiant_player_4_item3,radiant_player_4_item4,radiant_player_4_item5,dire_player_0_kills,dire_player_0_death,dire_player_0_assists,dire_player_0_last_hits,dire_player_0_gold,dire_player_0_level,dire_player_0_gold_per_min,dire_player_0_xp_per_min,dire_player_0_net_worth,dire_player_0_item0,dire_player_0_item1,dire_player_0_item2,dire_player_0_item3,dire_player_0_item4,dire_player_0_item5,dire_player_1_kills,dire_player_1_death,dire_player_1_assists,dire_player_1_last_hits,dire_player_1_gold,dire_player_1_level,dire_player_1_gold_per_min,dire_player_1_xp_per_min,dire_player_1_net_worth,dire_player_1_item0,dire_player_1_item1,dire_player_1_item2,dire_player_1_item3,dire_player_1_item4,dire_player_1_item5,dire_player_2_kills,dire_player_2_death,dire_player_2_assists,dire_player_2_last_hits,dire_player_2_gold,dire_player_2_level,dire_player_2_gold_per_min,dire_player_2_xp_per_min,dire_player_2_net_worth,dire_player_2_item0,dire_player_2_item1,dire_player_2_item2,dire_player_2_item3,dire_player_2_item4,dire_player_2_item5,dire_player_3_kills,dire_player_3_death,dire_player_3_assists,dire_player_3_last_hits,dire_player_3_gold,dire_player_3_level,dire_player_3_gold_per_min,dire_player_3_xp_per_min,dire_player_3_net_worth,dire_player_3_item0,dire_player_3_item1,dire_player_3_item2,dire_player_3_item3,dire_player_3_item4,dire_player_3_item5,dire_player_4_kills,dire_player_4_death,dire_player_4_assists,dire_player_4_last_hits,dire_player_4_gold,dire_player_4_level,dire_player_4_gold_per_min,dire_player_4_xp_per_min,dire_player_4_net_worth,dire_player_4_item0,dire_player_4_item1,dire_player_4_item2,dire_player_4_item3,dire_player_4_item4,dire_player_4_item5,winner
Data CART C4.5 Extra Trees Classifier Gradient Boosting Hist Gradient Boosting Random Forest Adaboost
~38k samples (<10min) 68.6 70.2 91.4 72.5 79.9 85.6 62.7
~33k samples (10-20min) 81.9 82.1 97.3 83.0 90.0 96.1 66.4
~42k samples (>20min) 92.5 91.9 98.6 90.8 96.1 98.6 67.8

Feature Reduction and Model Improvement

Selected Features: 46 Match Duration Tower States Barracks States Team Differences in Key Statistics

duration,radiant_series_wins,dire_series_wins,score,0_rts,1_rts,2_rts,3_rts,4_rts,5_rts,6_rts,7_rts,8_rts,9_rts,10_rts,0_dts,1_dts,2_dts,3_dts,4_dts,5_dts,6_dts,7_dts,8_dts,9_dts,10_dts,0_rbs,1_rbs,2_rbs,3_rbs,4_rbs,5_rbs,0_dbs,1_dbs,2_dbs,3_dbs,4_dbs,5_dbs,net_worth,assissts,last_hits,gold,level,gpm,xpm,winner

30k-40k samples

Data CART C4.5 Extra Trees Classifier Gradient Boosting Hist Gradient Boosting Random Forest Adaboost
~39k samples (<10min) 58.2 57.8 65.7 63.2 64.5 63.7 64.3
~33k samples (10-20min) 66.8 67.2 79.9 71.8 72.9 76.2 69.9
~42k samples (>20min) 77.5 78.6 91.1 77.5 80.7 86.6 72.2

47k-63k samples

Data CART C4.5 Extra Trees Classifier Gradient Boosting Hist Gradient Boosting Random Forest Adaboost
~56k samples (<10min) 58.1 58.3 64.6 63.4 64.5 64.4 63.9
~48k samples (10-20min) 65.1 66.3 78.1 69.6 72.6 75.4 70.3
~63k samples (>20min) 76.9 77.3 90.8 75.8 78.6 85.3 72.1

72k-98k samples

Data CART C4.5 Extra Trees Classifier Gradient Boosting Hist Gradient Boosting Random Forest Adaboost
~83k samples (<10min) 57.2 56.7 64.3 63.6 64.5 63.9 64.4
~72k samples (10-20min) 65.0 64.8 77.4 70.8 71.6 75.0 70.5
~98k samples (>20min) 74.8 74.5 89.6 75.2 77.3 84.7 72.8

Match Analysis with Hero and Item Embeddings

Selected Features: 50 duration,radiant_series_wins,dire_series_wins,score,0_rts,1_rts,2_rts,3_rts,4_rts,5_rts,6_rts,7_rts,8_rts,9_rts,10_rts,0_dts,1_dts,2_dts,3_dts,4_dts,5_dts,6_dts,7_dts,8_dts,9_dts,10_dts,0_rbs,1_rbs,2_rbs,3_rbs,4_rbs,5_rbs,0_dbs,1_dbs,2_dbs,3_dbs,4_dbs,5_dbs,radiant_item_embeddings,dire_item_embeddings,radiant_embeddings,dire_embeddings,net_worth,assissts,last_hits,gold,level,gpm,xpm,winner

Without Shuffle of Hero Embeddings

Data CART C4.5 Extra Trees Classifier Gradient Boosting Hist Gradient Boosting Random Forest Adaboost
~83k samples (<10min) 72.3 72.2 94.6 77.7 85.5 84.4 63.8
~72k samples (10-20min) 87.4 88.6 94.1 83.6 88.5 93.0 70.3
~98k samples (>20min) 92.8 93.3 97.9 88.2 93.2 97.6 72.5

With Shuffle of Hero Embeddings

Data CART C4.5 Extra Trees Classifier Gradient Boosting Hist Gradient Boosting Random Forest Adaboost
~83k samples (<10min) 58.0 57.9 65.1 63.6 64.9 63.9 63.8
~72k samples (10-20min) 64.6 64.0 74.0 70.4 71.8 72.4 69.8
~98k samples (>20min) 72.5 71.7 85.2 76.3 77.8 80.7 72.6

With Shuffle of Winner Hero Embeddings

Data CART C4.5 Extra Trees Classifier Gradient Boosting Hist Gradient Boosting Random Forest Adaboost
~83k samples (<10min) 57.6 58.0 65.7 64.1 65.3 64.5 64.0
~72k samples (10-20min) 64.3 64.5 75.7 70.5 71.8 73.4 69.8
~98k samples (>20min) 72.1 71.7 86.3 76.0 77.7 81.0 72.6

With Mean of Winner Hero Embeddings

Data CART C4.5 Extra Trees Classifier Gradient Boosting Hist Gradient Boosting Random Forest Adaboost
~83k samples (<10min) 62.0 61.3 68.6 70.2 73.1 68.0 63.8
~72k samples (10-20min) 75.2 75.7 81.8 76.8 78.1 79.6 70.3
~98k samples (>20min) 83.7 85.5 93.3 82.3 84.4 89.9 72.5

Hero Embeddings (Mean) + Item Embeddings (Mean)

Data CART C4.5 Extra Trees Classifier Gradient Boosting Hist Gradient Boosting Random Forest Adaboost
~83k samples (<10min) 60.7 62.5 69.4 71.4 72.6 68.6 64.5
~72k samples (10-20min) 73.8 74.0 83.2 77.3 78.4 79.9 70.9
~98k samples (>20min) 83.0 84.0 93.2 82.7 84.4 89.8 71.7

Hero Embeddings (Mean) + Item Embeddings (Mean) + more samples

Data CART C4.5 Extra Trees Classifier Gradient Boosting Hist Gradient Boosting Random Forest Adaboost
~96k samples (<10min) 60.9 61.5 68.5 70.2 71.6 67.6 64.6
~85k samples (10-20min) 73.8 76.5 82.4 75.7 76.9 80.0 71.0
~116k samples (>20min) 83.9 83.5 93.3 81.8 83.1 90.3 72.8

Hero Embeddings (Mean) + Item Embeddings (separate for each item, 108 features) + more samples

Data CART C4.5 Extra Trees Classifier Gradient Boosting Hist Gradient Boosting Random Forest Adaboost
~96k samples (<10min) 59.2 58.9 64.6 69.2 70.0 64.3 64.4
~85k samples (10-20min) 67.8 68.8 72.1 75.3 74.7 70.9 70.3
~116k samples (>20min) 77.1 78.5 81.3 82.0 82.2 78.1 73.0

Without embeddings

Data CART C4.5 Extra Trees Classifier Gradient Boosting Hist Gradient Boosting Random Forest Adaboost
~96k samples (<10min) 57.7 57.7 64.7 64.3 65.0 64.0 64.6
~85k samples (10-20min) 65.2 65.7 76.9 70.7 71.9 74.8 71.0
~116k samples (>20min) 75.7 74.9 89.6 76.1 77.7 85.0 72.8

Enhancing Predictive Models with Dotabuff Data: Hero and Item Winrate Features

Separate Hero Winrates for Teams

Data CART C4.5 Extra Trees Classifier Gradient Boosting Hist Gradient Boosting Random Forest Adaboost
~96k samples (<10min) 64.3 63.6 69.6 71.5 73.4 68.6 65.4
~85k samples (10-20min) 77.0 77.5 81.7 76.4 76.8 79.1 69.5
~116k samples (>20min) 86.0 87.1 92.8 82.5 83.9 90.3 72.3

Hero Winrate Difference

Data CART C4.5 Extra Trees Classifier Gradient Boosting Hist Gradient Boosting Random Forest Adaboost
~96k samples (<10min) 60.7 59.9 67.6 69.0 68.2 66.4 65.4
~85k samples (10-20min) 71.8 73.4 79.4 74.5 74.3 77.1 69.5
~116k samples (>20min) 82.5 82.6 91.2 81.0 80.4 88.1 72.3

Hero Winrate Difference + Item Winrate Difference

Data CART C4.5 Extra Trees Classifier Gradient Boosting Hist Gradient Boosting Random Forest Adaboost
~96k samples (<10min) 60.1 60.4 67.4 68.8 67.9 66.9 65.4
~85k samples (10-20min) 71.3 72.7 78.9 74.1 73.7 76.0 69.5
~116k samples (>20min) 81.4 81.3 91.0 80.7 80.7 86.7 72.3

Hero Winrate Difference + Item Winrate Difference + more samples

Data CART C4.5 Extra Trees Classifier Gradient Boosting Hist Gradient Boosting Random Forest Adaboost
~118k samples (<10min) 59.7 59.2 66.6 67.6 66.3 66.2 64.0
~105k samples (10-20min) 70.1 71.9 78.5 73.9 73.2 75.6 70.0
~142k samples (>20min) 80.1 81.6 90.5 79.4 79.5 86.1 72.63

Hero Winrate Difference + Item Winrate Difference + more samples

Data CART C4.5 Extra Trees Classifier Gradient Boosting Hist Gradient Boosting Random Forest Adaboost
~141k samples (<10min) 59.0 59.6 66.4 66.7 66.5 66.0 63.8
~127k samples (10-20min) 69.8 71.4 78.2 73.0 72.6 75.6 70.0
~173k samples (>20min) 79.8 79.5 90.3 78.6 78.7 85.7 72.8

Hero Winrate Difference + Item Winrate Difference + more samples

Data CART C4.5 Extra Trees Classifier Gradient Boosting Hist Gradient Boosting Random Forest Adaboost
~161k samples (<10min) 58.9 58.6 65.6 65.8 65.7 64.8 63.3
~146k samples (10-20min) 69.8 71.1 79.0 74.1 73.6 76.2 70.9
~201k samples (>20min) 78.7 79.7 90.4 79.0 78.9 85.9 73.4

Hero Winrate Difference + Item Winrate Difference + more samples

Data CART C4.5 Extra Trees Classifier Gradient Boosting Hist Gradient Boosting Random Forest Adaboost
~301k samples (<10min) 58.5 58.8 66.4 66.4 66.3 65.4 64.7
~273k samples (10-20min) 68.3 68.9 77.4 72.2 72.1 75.1 70.9
~376k samples (>20min) 77.4 78.5 89.0 77.1 77.5 84.6 74.2

Predicting Match Winner in Timelines

XG (radiant, winner) vs Tundra (dire)

Dota Plus Prediction

dp

Extra Tree Classifier

etc

Random Forest

rf

Hist Gradient Boosting

hgb

Mean of [Extra Tree Classifier, Random Forest, Hist Gradient Boosting]

avg

All in one + Mean

et_rf_hgb_avg

XG (radiant, winner) vs Tundra (dire)

Dota Plus Prediction

dp

Extra Tree Classifier

etc

Random Forest

rf

Hist Gradient Boosting

hgb

Mean of [Extra Tree Classifier, Random Forest, Hist Gradient Boosting]

avg

All in one + Mean

et_rf_hgb_avg

About

Master's thesis on the topic “Application of decision tree models to solve real problems”

Resources

Readme
Activity

Stars

0 stars

Watchers

1 watching

Forks

0 forks
Report repository

Releases

No releases published

Packages

No packages published

Contributors 2

  •  
  •  

Languages