Hello everyone, from the release of the competition questions yesterday until now, I have finally completed a complete finished product paper of mathorcup.
D Thesis has 60 pages in total, 7 pages for some revision instructions, and 43 pages for the main text.
I didn't expect to do it for so long. The feature data of this question is even more than that of my previous big data competition. It took two hours for the reliability analysis and outlier processing of the first question alone. The weight determination is good to say, brainless random forest. But the second question is disgusting again. The data description given in the title is translated. Therefore, I checked the literature for a long time about the actual meaning of pole position and related variables, and finally determined the pole and market volume. . . A formula is given after the quantitative method drawing analysis. The third question is brainless statistics, and the fourth question has too much feature data. After processing, I changed many machine learning models, and finally gbdt has the highest accuracy. The fifth question is a surprise, because it completely integrates the models of my previous four questions, and it is a perfect closed loop. I am more happy to do this.
The reason why it is so long is because:
a lot of space in my thesis needs to be used to explain why I do this, basically it is to teach you how to do it step by step, and I also need to take care of everyone's level, so there will be some places that need to be written very It is cumbersome, some intermediate processes are shown in detail, and there are many forms, you can put them in the appendix yourself
Here, let me tell you how to do it. I may not explain it in detail enough due to limited energy. You can click the card at the bottom of this article to view the video explanation.
OK, here is my directory:
Summary:
First question:
Some QAR data has errors, and it is necessary to preprocess the data to remove the false and preserve the true, so as to reduce the impact of the wrong data on the research analysis. Please invite your team to conduct reliability research on the data quality in Annex 1, extract some key data items related to flight safety, and analyze their importance.
The first is pre-analysis to see which data needs to be preprocessed and which data has a higher probability of error. For example, the probability of categorical variables having error data is relatively small. Time data, etc., are impossible to make mistakes. :
Then do the reliability analysis:
Use the coefficient of variation to judge whether there is an outlier, and after confirming which data exists, perform outlier processing:
After the processing is completed, here I directly determine the weight value of the random forest.
And this random forest will play a big role in the fifth question. :
Second question:
During the entire flight process from takeoff to landing, the aircraft ensures flight safety through a series of flight controls, which mainly include roll control, pitch control, etc. At present, domestic airlines use over-limit monitoring of flight control actions. Although this monitoring method can quickly distinguish the state deviation of the aircraft, it can only tell the safety management personnel what happened, and cannot immediately find out the cause of the deviation. For this reason, the cause of this deviation can be analyzed through the process change of the joystick. According to Annex 1, please provide a reasonable quantitative description of the flight control.
For this question, the most important thing is to understand the meaning of the question.
What is pole position? The question did not give us the data. This graph essentially just explains to us which points we should quantify and describe.
Note that our task is to find variables related to manipulation through the data given in the title, and use this variable to quantify and describe the manipulation process.
Which two quantities? The topic has hints:
But we need further analysis to show that it is indeed these two variables.
Here we need to search for the actual meaning of these two variables. These two words cannot be found in Baidu. I have searched a lot of foreign language materials, and the final summary is as follows:
1. Pole position:
The word "rod position" is often used to describe the position or attitude of the control stick when the pilot is operating the aircraft, such as "push the control stick to the front", "keep the control stick in the neutral position" and so on. According to the requirements of the aircraft and flight operation procedures, the pilot adjusts the position of the joystick to control the movement of the aircraft and maintain the stability of the flight.
2. The volume affects the slope:
The roll of the aircraft refers to the movement of the aircraft rotating around its longitudinal axis , that is, the angle change of the left and right tilt. Pilots typically use the aircraft's joystick, or joystick, to control the aircraft's bank to achieve the desired roll motion. Rotation of the joystick or stick from side to side causes the aircraft's bank angle to change, which controls the aircraft's roll motion.
"Turn amount" is the amount of amplitude or angular change the pilot turns the joystick or stick while manipulating the aircraft. For example, when a pilot rotates a stick to control the roll motion of an aircraft, "dial volume" can be used to describe the angle of rotation of the stick . Similarly, when a pilot rotates a stick to control the heading of an aircraft, "dial volume" can also be used to describe how much the stick is turned.
3. Rod amount affects pitch angle:
"Stick amount" refers to the displacement amount of the joystick . When the pilot operates the joystick, he needs to push it back and forth a certain distance to achieve the desired flight control.
Look back at the topic again:
Perfect, exactly as defined by these two variables.
After that, since there is no pole position data, which points need to be manipulated and described by the G value:
Import the other two data and draw together to see how to describe it:
OK, the solution is complete.
The third question:
Please study the data in Appendix 2, analyze the different situations of overruns, and study the basic characteristics of different overruns, such as analyzing which routes the aircraft is on or which airports are prone to overruns, etc.
so tired. . . . Let me be brief and casual from this question onwards. . For details, you can watch the explanation video in my previous summary post.
The frequency analysis ends.
Fourth question:
Please build a mathematical model, explore a flight technology evaluation method based on flight parameters, and analyze the pilot's flying technology. The "different qualifications" in the data table represent the different skill levels of the pilots.
Machine learning trains a classification model and it's done:
Fifth question:
If you are the safety manager of the airline, please establish the airline's real-time automatic early warning mechanism to prevent possible safety accidents, and combine the data in Appendix 1 to give the simulation results.
I am very happy to do this question, because it perfectly verified the model of my previous four questions and formed a closed loop tightly. :
First question:
What is obtained is to judge some key data related to flight safety based on the aircraft data.
But please note that the key data here is obtained through random forest prediction, that is to say, what we finally obtain is a training model that judges the safety of the aircraft based on all the data of the aircraft.
Second question:
Quantitative indicators for evaluating whether to make a hard landing based on aircraft data.
The third question:
Statistical models of overrun types based on aircraft data.
Fourth question:
A Pilot Skill Assessment Model Based on Aircraft Data.
Needless to say, according to these four models, after importing aircraft data, what can be obtained and how to combine them to build an early warning framework model will have detailed answers in my thesis:
OK, let’s stop here. I’m too tired, and the explanation may not be detailed enough. For a detailed explanation, please move to my explanation video. It will be a detailed explanation:
For viewing the above complete finished product itself, you can click on the card below: