Based on https://github.com/Divergent914/kddcup2015

This leaderboard calculates scores by about 50% of the test data.

AUC score: 0.8874428893001793

Rank: #89

AUC score: 0.8867360110438458

Rank: #86

0. 加特征

1. 加数据：最多可以到depth=4

2. feature selection

3. 观察预测错了的instance

4. 尝试深度学习model，http://card.memect.com/?q=keras

5. kNN

6. GBDT: min_samples_leaf, max_depth ...

7. RF: min_samples_leaf, max_depth ...

8. AdaBoost(with LR/SVC)

9. 把测试集拿来一起做normalize

• 用户在该课程的操作数量（按event_source序对统计），enrollment最后一周、倒数第二周、第一周、总体，占该用户在所有课程的比例，占该课程所有用户的比例 (108)

• 用户有行为的课程数量，用户行为的最后一周、倒数第二周、第一周、总体 (4)

• 课程的选课人数 (1)

• trending slope of the weekly number of events within the enrollment (1)

• numbers of events in the last week, the first week, and the week before the last week, and all weeks of the enrollment; ratio on all courses by user; ratio on all users by course (12)

• average, standard deviation, maximal, minimal weekly numbers of events in the enrollment period (4)

• coefficients b and c in the polynomial model y = a + bx + cx**2, where x is week number (all start from 0), and y is the weekly number of events (2)

• 7 counts of events in Monday to Sunday, by event-source pairs, and ratio on total counts (203)

• 24 counts of events in hour 0-23, by event-source pairs, and ratio on total counts (696)

• 7 counts of event types, and ratio on total counts (14)

• 2 counts of source types, and ratio on total counts (4)

• number of 3-hour defined sessions in the enrollment (1)

• average, standard deviation, maximal, minimal numbers of events in 3-hour defined sessions in the enrollment (4)

• statistics of 3-hour defined sessions: mean, std, max, min of duration (4)

• number of 1-hour defined sessions in the enrollment (1)

• average, standard deviation, maximal, minimal numbers of events in 1-hour defined sessions in the enrollment (4)

• statistics of 1-hour defined sessions: mean, std, max, min of duration (4)

• number of 12-hour defined sessions in the enrollment (1)

• average, standard deviation, maximal, minimal numbers of events in 12-hour defined sessions in the enrollment (4)

• statistics of 12-hour defined sessions: mean, std, max, min of duration (4)

• number of 1-day defined sessions in the enrollment (1)

• average, standard deviation, maximal, minimal numbers of events in 1-day defined sessions in the enrollment (4)

• statistics of 1-day defined sessions: mean, std, max, min of duration (4)

• number of 7-day defined sessions in the enrollment (1)

• average, standard deviation, maximal, minimal numbers of events in 7-day defined sessions in the enrollment (4)

• statistics of 7-day defined sessions: mean, std, max, min of duration (4)

• 课程材料首次发布、最近发布距今几天 (2)

• 用户初次、上次操作此课程据今几天，持续几天，与课程持续时间的比例，初次访问课程材料距离开课时间几天 (5)

• 课程的所有用户操作课程持续时间的：平均值、标准差、最大值、最小值，以及与课程持续时间的比例 (8)

• month (1-12) of the first, last event in the enrollment (2)

• 用户对课程材料的首次操作时间与课程材料发布时间的日期差的：平均值、标准差、最大值、最小值，enrollment最后一周、倒数第二周、第一周、总体 (16)

• 课程最后十天有操作的人数，与选课人数的比例: 课程有更新的最后十天、课程有操作的最后十天 (3)

• 用户平均每个课程Dropout的次数，在该课程中Dropout的次数，与平均每个课程Dropout次数的比例；总Dropout持续时长与课程持续时间的比例 (4)

• 用户在所有课程上Dropout总持续时长与课程持续时间的比例的：平均值、方差、最大值、最小值；有Dropout行为的课程占用户所选课程总数的比例 (5)