Category: Uncategorized

Why does it sometimes not work?

Bias Appearing Reasons in the Forecasting Process

by Ali Jooyafar <jooyafar@gmail.com> on P41-42 of 12th Rahbord Mag. (Download)

Have you heard the story of the deaf man and his insistence on visiting his sick friend? a story by Maulana Jalaluddin Mohammad Balkhi. Do you remember that he predicted all possible dialogues before meeting the patient and had an answer in mind for them? When meeting with the patient, the patient’s first words turned out to be contrary to his imagination. The man thought to himself that the story was unfolding as he had predicted. One after another, the patient’s questions were further from the deaf’s imagination; But the deaf person gave the same preset answers. This process continued in a domino way and the further it went, the worse the situation was. In this article, we will discuss some reasons for the difference between predictions and reality.

The Necessity of Relative Knowledge of The Expected Field

In order to build a production or business group, it is necessary to anticipate its various aspects. The fact that brands go to the similar group for the development of their product or service group is the mastery of the knowledge of the production of the similar group of goods or services and the experience they have gained from operating in the same field. Brands that disobeyed this rule have mostly failed. A successful example of this is the Sabaideh group, which launched Filmo after the success of Aparat. Or the Tavakli brand, which, in addition to producing matches, turned to the production of chipboard and cabinets. As a negative example, we can also mention Sony and Google, which, due to their high share in the digital services market, felt that they could also enter the production of goods such as mobile phones, but they failed after a while. In the examples mentioned, success or failure depended on familiarity with the space that the person dared to enter. In the example at the beginning of the article, think if the deaf man was aware of the mood of his sick friend, would he ever make such predictions?

Get Feedback

Do you think that in the story of the deaf man, if there was a person who, with a delay of two minutes and using sign language, informed the deaf person about the situation in the environment, which was becoming more critical by the minute, would he still have continued that dialogue? Now think about the amount of this number, maybe announcing the report from moment to moment is costly for the system, and on the other hand, not being up-to-date for a long time and not being aware of the system’s conditions will raise other costs that are caused by wrong predictions. Therefore, in every system, there should be a measure for getting feedback in the first step, and in the second step, a suitable number for the distance between feedbacks should be determined.

The Worth of Older Data

Have you ever paid attention to predicting the computer’s time to perform an operation? First it writes: 1 hour left, then it becomes 6 hours, 1 day, 1 year, and then suddenly it goes back to 30 minutes and it’s over! This may be due to reliance on the most up-to-date data. For example, suppose you have issued an order to transfer 100 data, your first data will be transferred quickly due to its small size or less RAM memory usage; But the second file is transferred later due to other reasons. The computer also estimates that the speed of operations for other files is equal to the speed of the last (newest) transferred file. You might say to yourself that it is better for the computer to exclude the data transfer speed that is transferred too fast or too slow from the calculations and take an average from the other data. We call this method the “simple average” method, which is the first forecasting method. If you say that only the last 3 data can be averaged: the “3-period moving average” method, and if you use the same method with the difference that based on the new/old data, such as calculating the average of your courses, you will give them a coefficient: “Weighted moving average 3-periodic”, in fact you allow any data that has occurred to play a role in predicting a new data as close as it is to it. In an era, if a black child was born from white parents, the mother was suspected; While later they said, for example, that his ancestor was black (and by chance, it caused his ancestor’s gene to take a higher factor here and become black). In this case, it can be said that the weight of the data was not based on the proximity to the new data, but it was designed as a random number (factor), which may, in a few examples, suddenly create a large effect (a high coefficient) on the new generation, or the sum of the coefficients. Generations that have been black have become more than the sum of the coefficients of other generations. Now consider examples where each event is independent of the other, and the probability of each occurrence is very low or high; Here, it is better to allocate less probability to the nearer periods! To better understand this issue, I ask a question:

What do you think is the safest flight line? Yes! As the last flight has crashed; But thinking of the previous subject, people avoid using the services of that service provider for a long time after the occurrence of a disaster!

Data Verification

Have you noticed that to show the level of trust in a person, they say that they trust him more than their eyes or ears? In fact, the eyes and ears are much more worthy than any other person or source as a source that gives real-time feedback; But what does this more or even equal trust mean? Is it just a compliment and exaggeration? No! This trust comes from the infallibility of the information coming from a source, and has nothing to do with the speed or up-to-dateness of the data. In fact, in the story of the deaf man, it can be assumed that he was not deprived of moment-by-moment reporting; But this moment-to-moment report of his was from an imaginary source and based on his own speculation, which was giving him false information.

Such a claim was also made in the accident of firing 2 missiles at Ukrainian flight PS752; On this flight, incorrect radar data due to incorrect settings (and some gross user errors: see Swiss Cheese Model) killed 176 people.

Data Correction

Suppose the error of your data (the difference between the actual value and the prediction you had) after a few periods is equal to: +20, +10, 0, -10, -20, -10, 0, +10, +20, + 10, …

Notice the error pattern? Your error follows a pattern and repeats itself once every 8 cycles. In this example, the method we have chosen for forecasting is regular and applies the same error to 8k+i periods, and the forecast can be modified at each stage. For example, if 240 numbers are predicted in period 132; Because 8×16+4=132, then our model has predicted it with an error of -10, and the true answer is equal to: 240-(-10)=250

Attention to The Outer Factor

Further, to predict the real demand, after determining the normal demand forecast at a certain time, which is dependent on the passage of time, a number around the number 1 is also multiplied by that normal value, so that the changes related to that season or period of time are applied. We call this factor the seasonal factor that depends on “customer needs”. For example, at the end of the year, the demand for detergents increases, and this increase depends on time. Also, in case of advertising by the marketing team, it may be predicted that the demand will receive another coefficient based on “customer demand”, which will lead to an increase in demand; For example, let’s assume that in order to predict the demand of the cooling product group in the winter season, which is 0.75 times, we will predict the sales of this product by 1.5 times with advertising and marketing (for example, winter sales discount). The sale of the product is 200 units from the first day of sale, and 30 units are added to it at the beginning of the following seasons. We have started production since the fall of last year. The demand calculation for this product is as follows: 394=1.5×0.75×(200+5×30)

Being Multivariate

Many simulation problems do not accurately reproduce the output of the problem after re-run even with the same conditions with the same user, same location, same settings, etc.; But these outputs are very close to each other, and the reason for that is the involvement of random numbers in the models, which causes these close answers to be unacceptable only in sensitive medical and military cases, and acceptable in other cases; But the point is that by changing any of the main variables, such as the number of users, system regulations, executive rules, etc., the final answer will change significantly. A good forecaster always tries to identify each factor affecting the final answer and measure the effect of each one with the Ceteris Paribus rule (refer to: Ceteris Paribus).

References

– Andersen JR, Byrjalsen I, Bihlet A; et al. (2015). “Impact of source data verification”. Br J Clin Pharmacol. 79 (4): 660–8.-

رزمی، جعفر و لطفی، محمدمهدی (1396). اصول برنامه‌ریزی تولید و کنترل موجودی‌ها. تهران: انتشارات دانشگاه تهران.

– علینقیان، مهدی؛ ایزدبخش، حمیدرضا و زرین‌بال، معصومه (1393). مقدمه‌ای بر شبیه‌سازی سیستم‌های گسسته – پیشامد. تهران: انتشارات موجک.

– دفتر بررسی سوانح و حوادث هوایی ج. ا. ایران (1399). گزارش نهایی بررسی سانحۀ پرواز PS752.

-صادقیه، احمد (1384). تصمیم‌گیری براساس الگوریتم ژنتیک در بهینه‌سازی. یزد: انتشارات علم نوین.