Development of a Program or Model for Analyzing Existing Predictive Data
Budget: €750 – €1,500 EUR
Hello!
I am seeking to commission the creation of a program or machine learning model that analyzes predictive data against actual data.
Data for Analysis:
The program will analyze data contained in a .csv file with the following structure:
1. Hourly timestamp in column 1.
2. Actual hourly financial data values in column 2.
3. Predicted financial data values for the next 12 hours, spread across columns 3-14.
Objective:
The goal is to determine the most probable moments indicating a trend change. The model should maximize the net positive outcome by minimizing the total value of incorrectly predicted trend changes subtracted from the total value of correctly predicted changes. Ideally, the dependencies used to predict should be minimal (preferably one universal, or at most three non-conflicting), and should be applicable to all suitable cases to maximize the potential positive outcome after accounting for losses incurred by the algorithm/model use.
Quantitative Goal Definition:
The objective for the algorithm or model is to accurately identify directional changes from its algorithm-defined pivot point to the next change indication. The process involves summing all correct change values in the dataset and subtracting the sum of all incorrect values determined by the chosen method for identifying pivot points. The optimal method for determining pivot points should yield the highest possible net positive value of correctly identified changes versus incorrectly identified ones. For clarification, refer to Figure 1 attached.
Order Components:
-Development of an optimal algorithm or model that marks pivot points.
-The algorithm or model should be capable of retraining in the future with a larger historical data set for optimization tests.
-The trained program or model must have the capability to:
I. Marking in the data table:
-In an additional 15th column, mark the presumed pivot points for an uptrend by placing a value at the proposed change moment, in the row with the corresponding timestamp indicating when this reversal will occur,
-In an additional 16th column, mark the presumed pivot points for a downtrend by placing a value at the proposed change moment, in the row with the corresponding timestamp indicating when this reversal will occur. Please refer to the illustrative Figure 2.
II. The ready-to-run program should analyze data every hour, 90 seconds after each full hour, based on a 24-hour window (.csv file with 12 hours of history and 12 hours of prediction), erase previous points from the .csv file of that window, and update it with new, updated points (thus avoiding leaving points that might be deemed incorrect with more analyzed data).
Program Preparation:
a) Initial preparation of the program and marking of data in the file provided for analysis (the file has about 600 hours/rows with predictions).
b) Upon acceptance of the first phase, completion of the program, i.e., adding the functionalities discussed above enabling its continuous operation on new data.
Suggestions:
-Since the prediction lines are 12 hours long and partially overlap every hour, a repeating pattern of directional change in several consecutive predictions or in most of a specific group (e.g., three out of the nearest four predictions) is a starting point for decision-making. However, it remains to be mathematically determined whether the most recent predictions are most valuable as they approach a pivot point. It is worth testing whether, if the predictive data consistently indicates an approaching pivot point, we should average these indications as shown in Figure 3a, or assign greater value to the most recent predictions and shift emphasis toward them as shown in part b) of the same figure.
-We should also determine whether the angle of inclination of curves formed from predictive points is significant and, if so, to what extent.
-As I am not a data scientist, I am also open to other suggestions and insights from experienced individuals that could optimize the program's performance.
I am seeking to commission the creation of a program or machine learning model that analyzes predictive data against actual data.
Data for Analysis:
The program will analyze data contained in a .csv file with the following structure:
1. Hourly timestamp in column 1.
2. Actual hourly financial data values in column 2.
3. Predicted financial data values for the next 12 hours, spread across columns 3-14.
Objective:
The goal is to determine the most probable moments indicating a trend change. The model should maximize the net positive outcome by minimizing the total value of incorrectly predicted trend changes subtracted from the total value of correctly predicted changes. Ideally, the dependencies used to predict should be minimal (preferably one universal, or at most three non-conflicting), and should be applicable to all suitable cases to maximize the potential positive outcome after accounting for losses incurred by the algorithm/model use.
Quantitative Goal Definition:
The objective for the algorithm or model is to accurately identify directional changes from its algorithm-defined pivot point to the next change indication. The process involves summing all correct change values in the dataset and subtracting the sum of all incorrect values determined by the chosen method for identifying pivot points. The optimal method for determining pivot points should yield the highest possible net positive value of correctly identified changes versus incorrectly identified ones. For clarification, refer to Figure 1 attached.
Order Components:
-Development of an optimal algorithm or model that marks pivot points.
-The algorithm or model should be capable of retraining in the future with a larger historical data set for optimization tests.
-The trained program or model must have the capability to:
I. Marking in the data table:
-In an additional 15th column, mark the presumed pivot points for an uptrend by placing a value at the proposed change moment, in the row with the corresponding timestamp indicating when this reversal will occur,
-In an additional 16th column, mark the presumed pivot points for a downtrend by placing a value at the proposed change moment, in the row with the corresponding timestamp indicating when this reversal will occur. Please refer to the illustrative Figure 2.
II. The ready-to-run program should analyze data every hour, 90 seconds after each full hour, based on a 24-hour window (.csv file with 12 hours of history and 12 hours of prediction), erase previous points from the .csv file of that window, and update it with new, updated points (thus avoiding leaving points that might be deemed incorrect with more analyzed data).
Program Preparation:
a) Initial preparation of the program and marking of data in the file provided for analysis (the file has about 600 hours/rows with predictions).
b) Upon acceptance of the first phase, completion of the program, i.e., adding the functionalities discussed above enabling its continuous operation on new data.
Suggestions:
-Since the prediction lines are 12 hours long and partially overlap every hour, a repeating pattern of directional change in several consecutive predictions or in most of a specific group (e.g., three out of the nearest four predictions) is a starting point for decision-making. However, it remains to be mathematically determined whether the most recent predictions are most valuable as they approach a pivot point. It is worth testing whether, if the predictive data consistently indicates an approaching pivot point, we should average these indications as shown in Figure 3a, or assign greater value to the most recent predictions and shift emphasis toward them as shown in part b) of the same figure.
-We should also determine whether the angle of inclination of curves formed from predictive points is significant and, if so, to what extent.
-As I am not a data scientist, I am also open to other suggestions and insights from experienced individuals that could optimize the program's performance.