Survival Rate Prediction Model for Startup Companies

This is an end semester project that I have done with my groupmates in college. In this study, we proposed a model to predict a startup company’s future condition using a deep multilayer perceptron (MLP) and decision tree. I mainly played the role for data quantification and literature review of this study. In our study, we built a prediction analysis model for startup companies. Furthermore, we identified what the key factors are and what influences will it have on the results. Our main hypothesis is that money, people and active days are the key factors. We built the prediction model from CrunchBase, which is the largest public database with relevant profiles about companies.

Data Quantification

For the data obtained, there are 4 groups of data that are quantified: country state, employee range, roles and countries. Regarding the states group (which only exists in USA and Canada), we think that different states have different impacts on the survival rate of a startup. A lookup table for scoring of states is defined [1][2][3]. The original data for employee quantities are a range between two numbers (e.g. 101-250). These are transformed into the average of the upper and lower bound. Employees of 10000+ are defined to be 15000. For the roles group, the data “company” is arbitrarily defined as 0.1 and “company, investor” is defined as 0.9. This is because we regard a company as wealthier and influential when it also plays a role of an investor, compared with only being a company. The other roles are transformed to 0.5. The impact of country on startup environment is also studied [4] and scores from 0.329 to 0.947 are given to countries that have above 300 startup companies in record.

Table 1. Company status and scores of 23 countries.

Country Closed Operating Acquired IPO  Score 
BRA4.7%90.1%4.5%0.6%0.329
ESP5.3%86.8%7.3%0.7%0.333
FIN3.6%82.9%12.6%0.9%0.340
ITA2.9%89.5%6.6%1.0%0.345
RUS6.0%87.5%4.7%1.8%0.379
KOR4.9%89.0%3.5%2.6%0.411
BEL4.3%80.6%12.3%2.8%0.419
NLD3.6%83.5%10.1%2.8%0.421
GBR5.2%81.4%10.5%2.9%0.423
DEU5.5%78.7%12.9%2.9%0.423
IRL5.8%78.9%12.0%3.3%0.443
DNK4.4%79.6%12.1%3.9%0.466
CHE2.7%83.3%10.0%4.0%0.471
SWE3.9%82.3%9.8%4.0%0.473
USA8.1%69.3%18.5%4.2%0.478
IND3.3%85.5%6.8%4.4%0.487
SGP3.2%85.5%6.5%4.8%0.507
FRA4.0%78.4%12.6%5.1%0.516
ISR5.4%74.7%13.7%6.2%0.563
JPN3.4%82.2%5.4%9.0%0.684
CAN5.0%70.6%13.3%11.1%0.769
CHN3.3%79.2%4.0%13.5%0.874
AUS3.3%77.1%4.4%15.3%0.947

Neural Network Implementation

The ANN runs on a Windows 10 OS (i7-8700k CPU) with DDR4 2666MHz 16G RAM and Nvidia GTX 1070 Ti GPU. Keras is used to construct the network, and three networks of multilayer perceptron (MLP) with different number of layers are designed for comparison. There are 4 outputs of the network, indicating the probabilities of the final status which are: (1) Closed, (2) Operating, (3) Acquired and (4) IPO.

Figure 1. MLP Structure of the 3 neural networks.

The three networks have respectively 2, 4 and 6 hidden layers, with each network all starting with a 1024-neuron hidden layer and ending with a 32-neuron hidden layer (Fig. 1). Table 2 shows the training results.

Table 2. Mean square error (MSE) and categorical cross-entropy loss for the 3 networks.

LossActivation Network 1 Network 2 Network 3 
MSEReLU0.6950.6940.689
MSEsigmoid0.6900.6900.691
MSElinear0.6920.6890.692
MSEtanh0.6880.6930.691
Cross-entropy loss ReLU0.6950.6910.690
Cross-entropy losssigmoid0.6920.6910.691
Cross-entropy losslinear0.6950.6920.690
Cross-entropy losstanh0.6910.6920.691

The results show that almost all results are close to 70% with few difference. In general, ReLU activation has a better result than sigmoid activation, while linear activation may outperform ReLU when the network is deep enough.

The neural network behaves like a black box. It is quite difficult to conclude significant insights just by looking at the trained parameters. However, regarding the importance for business, we used a decision tree to help us understand which factor is the most important and how important they are respectively.

Decision Tree Training

We tried three different depths of decision tree: 4 (Fig. 2), 6 (Fig. 3), and 10. We set the gain ratio to be the criterion, and the confidence to 0.1.

Figure 2. Decision tree of depth = 4.

Figure 3. Decision tree of depth = 6.

The results show that if a company is large enough to exceed 500 people, the close rates are low. In most cases, large companies are acquired by mergers and acquisitions. In addition to the number of employees, funding total amount also has a significant impact on whether a company eventually survives or closes.

Key Findings

Among the factors regarding the ability to survive of starting up companies, the factors such as the number of employees, funding total amount, and the active days have significant influences on the company’s survivability. On the contrary, the factors such as country, region or number of funding rounds do not have significant influences. Whether a company acts as an investor simultaneously will also have influences on whether the company will become an IPO or will be acquired.

Future Work

Our future work is expected to integrate the inspirative insight gained from our case study with the methodology of our own. Three items are listed as in the following:

  1. Construct a heterogeneous relationship network for survival rate prediction [5].
  2. Define a data path score according to HeteSim algorithm [6][7].
  3. Predict company survival rate using MLP, decision tree and other neural networks.
  4. Predict how much money a company will raise.

References

  1. Bill Murphy , The Start-up Hall of Shame (America’s 10 Worst States for Entrepreneurs), © 2018 Manuseto Ventures, inc.com/bill-murphy-jr/the-startup-hall-of-shame-americas-10-worst-states-for-entrepreneurs.html
  2. Bill Murphy , 10 Top States for Entrepreneurship and Innovation, © 2018 Manuseto Ventures, inc.com/bill-murphy-jr/ranking-the-10-top-states-for-entrepreneurship-and-innovation.html
  3. Enterprising States: States Innovate, © 2015 The U.S. Chamber of Commerce Foundation, www.uschamberfoundation.org/enterprisingstates/
    assets/files/Executive-Summary-OL.pdf
  4. Zameena Mejia, The top 10 best countries for entrepreneurs in 2018, © 2019 CNBC LLC,
    https://www.cnbc.com/2018/02/05/
    us-world-news-report-2018-top-10-best-countries-for-entrepreneurs.html
  5. Xiangxiang Zeng, You Li, Stephen C.H. Leung, Ziyu Lin, Xiangrong Liu, Investment behavior prediction in heterogeneous information network, Neurocomputing, Volume 217, 2016, Pages 125-132
  6. Sun, Y., & Han, J. (2012). Mining Heterogeneous Information Networks: Principles and Methodologies. Synthesis Lectures on Data Mining and Knowledge Discovery, 3(2), 1-159
  7. Shi, C., Kong, X., Huang, Y., Yu, P. S., & Wu, B. (2014). HeteSim: A General Framework for Relevance Measure in Heterogeneous Networks. IEEE Transactions on Knowledge and Data Engineering, 26(10), 2479-2492. [6702458].

Coffee Maker Alarm

This is a project which I and my classmates had finished in a course focusing on practical implementation for mechatronics and system design, and is a sequel of the previous Aroma Alarm Clock project. We designed an Android app-controllable alarm clock which can heat up objects such as a water-filled container, and intend to further improve it for making coffee, so that one can enjoy a fresh cup of coffee in the morning after awakening. Similar to the previous one, I worked as the engineer in our group, and designed and constructed the hardware system.

System Framework

Fig. 1 shows the overall system architecture, which several electronic components (e.g. clock module, LCD display…) are controlled by a microcontroller that can communicate with a mobile application using a Bluetooth module. The alarm clock can be controlled either using its own hardware input (buttons) or using the mobile app. Simply put, the system has same the functions as a typical alarm clock, but has two additional features:

1. Temperature controllable heater.
2. Bluetooth communication.

Figure 1. System architecture for the coffee maker alarm.

Hardware

Arduino Uno, DS1302 chip, HC-06 and LCD1602 are used respectively for the microcontroller, real-time clock module, Bluetooth module and LCD display module. Being an end-semester project in class, the components are simply connected using a breadboard. All the hardware components except the heater are packed using patterned cardboard in a simply looking fashion (Fig. 2).

Figure 2. Hardware appearance of the coffee maker alarm.

A ceramic heater plate and temperature sensor are integrated for realizing temperature controlling (Fig. 3), and are connected out from the main hardware. The total cost for all the components is 1,741 NTD (≅ 60 USD).

Figure 3. Heater and temperature sensor.

Software

App Inventor 2 is used as the platform for creating the Android app for controlling the hardware. Fig. 4 displays the designing interface for the mobile app using the platform.

Figure 4. Design interface using App Inventor 2.

The coding section for this platform is very unique and easy to get started. It implements a block-based programming method which developers can create procedure by dragging pre-defined blocks together to for performing a certain function. Developers with no programming background can use this kind of environment for creating mobile apps. Fig. 5 shows a gallery containing the complete code for the alarm clock controlling app.

Figure 5. Gallery of block codes for the alarm clock app.

Here’s a video demonstration for using this system:

After finishing this project, I acquired important skills for system design and mechatronics integration. This enabled me to create more interesting and sophisticated projects such as the Automated Microfluidic Controlling Platform, Remote Commandable Self-Driving Toy Car and Real-time Impedance Detection Systems.

Reinforcement Learning applied to Forex Trading

It is already well-known that in 2016, the computer program AlphaGo became the first Go AI to beat a world champion Go player in a five-game match. AlphaGo utilizes a combination of reinforcement learning and Monte Carlo tree search algorithm, enabling it to play against itself and for self-training. This no doubt inspired numerous people around the world, including me. After constructing the automated forex trading system, I decided to implement reinforcement learning for the trading model and acquire real-time self-adaptive ability to the forex environment.

Environment Setup

The model runs on a Windows 10 OS (i9-9900K CPU) with DDR4 2666MHz 16G RAM and NVIDIA GeForce RTX 2060 GPU. Tensorflow is used for constructing the artificial neural network (ANN), and a multilayer perceptron (MLP) is used. The code is modified from the Frozen-Lake example of reinforcement learning using Q-Networks. The model training process follows the Q-learning algorithm (off-policy TD control), which is illustrated in Fig. 1.

Figure 1. Algorithm for Q-learning and the agent-environment interaction in a Markov decision process (MDP) [1].

For each step, the agent first observes the current state, feeds the state values into the MLP and outputs an action that is estimated to attain the highest reward, performs that action on the environment, and fetches the true reward for correcting its parameters. The agent follows the epsilon-greedy policy (ε = 0.1) for striking a balance between exploration and exploitation.

State, Action and Reward

For the 1st generation, price values at certain time points and technical indicators are used for constructing the states. The technical indicators used are the exponential moving average (EMA) and Bollinger bands (N=20, k=2), and time frames of 1, 5 and 15min are used with the last 10 time points being recorded. A total number of 36 inputs are connected to the MLP.

There are three action values for the agent: buy, sell and do nothing. The action being taken by the agent is determined by the corresponding three outputs of the MLP, where sigmoid activation functions are used for mapping the outputs to a value range of 0 ~ 1, representing the probability of the agent taking that action.

For the reward function, the difference between the trade price (the price when a buy/sell action is taken) and the averaged future price is considered. If a buy action is taken, then the reward function is calculated by subtracting the averaged future price with the trade price; if a sell action is taken then the reward is calculated the other way around. For “do nothing” actions, the reward is 0. A spread is subtracted from the reward for buy/sell actions to obtain the final reward. This prevents the agent to perform actions that result in insignificant profit, which would likely lead to a loss for real trades (Fig. 2).

Figure 2. Reward calculation method for buy/sell actions.

Noisy Sine Function Test

For preliminary verification of effectiveness for the training model and methods, a noisy sine wave is generated with Brownian motion of offset and distortion in frequency. This means at a certain time point (min), the price is determined by the following equation:

$$P(t)=P_{bias} + P_{amp} sin{2\pi \over T}t+P_{noise}$$

where Pbias is an offset value with Brownian motion, Pamp is the price vibration amplitude, T is the period with fluctuating values, and Pnoise is the noise of the price with randomly generated values. (Note that the “price” mentioned here is defined as the exchange rate between two currencies)

Fig. 3 shows a randomly generated price vs time sequence within a range of 50,000 minutes with an initial values Pbias = 1.0, T = 120 min, Pamp = 0.005, and Pnoise amplitude = 0.001. Generally, the price seems to fluctuate randomly with no obvious highs or lows. However, if it is viewed close-up, waves with clear highs and lows can be observed (Fig. 4).

Figure 3. Price vs time of the noisy sine wave from 0 to 50,000 min.

Figure 4. Price vs time of the noisy sine wave from 20000 to 20600 min.

The whole time period is 1,000,000 min (approximately 700 days, or 2 years). Initially, a random time period is set for the environment. Every time the agent takes an action, there is a certain chance (= 1%) that the time will jump to another random point within the whole period. Otherwise, the time will move on to a random point which is around 1 ~ 2 day(s) in the future. This setting is expected to correspond to real conditions, where a profitable strategy can have stable earnings and can also adapt quickly to rapid changing environments.

Fig. 5 plots the cumulative profit for trading using the noisy sine wave signal for 50,000 steps. Although it took approximately 25,000 steps to make the model get “on track”, I recognize this result as an important start for implementing real data.

Figure 5. Cumulative profit from trading using a noisy sine wave signal.

Fundamental Analysis for Economic Events

Fundamental analysis is a tricky part in forex trading, since economic events not only correlate with each other, but also might have opposite effects on the price at different conditions. In this project, I extracted the events that are considered significant, and contain previous, forecast and actual values for analysis. Data from 14 countries of the past 10 years are downloaded and columns with incomplete values are abandoned, making a complete table of economic events.

Because different events have different impacts on forex, the price change after the occurrence of an event is monitored, and a correlation between each event and the seven major pairs (commodity pairs). Table 1 displays a portion of the correlation table for different economic events. The values are positive, which indicates the significance of an event on the currency pair. Here, a pair is denoted by the currency other than the USD (e.g. USD/JPY is denoted as JPY).

Table 1. Correlation table between 14 events and 5 currency pairs. Here, a pair is abbreviated as the currency other than the USD.

Country Economic Event (Index)AUDCADEURGBPJPY
AUDCommodity Prices0.00313 0.00268 0.00266 0.00339 0.00278
AUDMI Inflation Expectations0.003380.001680.002170.002000.00266
AUDRBA Interest Rate Decision0.004280.002620.002580.002980.00225
EURManufacturing PMI0.003310.002840.002630.002980.00278
EURItalian CPI0.003150.003190.002950.003160.00255
EURServices PMI0.003410.002900.002930.002950.00284
EURCPI0.003040.002940.002620.003170.00241
EURGerman Unemployment Rate 0.003150.003150.002730.003130.00246
EURECB President Trichet Speaks0.003440.002480.003410.003020.00268
EURGerman Unemployment Change 0.003130.003130.002680.003070.00243
EURGerman Trade Balance0.003060.002550.003000.002840.00268
EURGerman Factory Orders0.002920.002650.003120.002800.00275
EURGerman Retail Sales0.003040.003040.003530.003100.00275
EURFrench Trade Balance0.003120.003120.002960.003010.00299

A total of 983 events are analyzed. However, due to the fact that a large portion of events have little influence on the price, only 125 events that have a relatively significant impact are selected as the inputs of the MLP.

Real Data Implementation Results

Per-minute exchange rate data of the seven currency pair is downloaded from histdata.com. A period from 2010 to 2019 is extracted, and blank values are filled by interpolation. This gives us a total of approximately 23 million records of price data (note that weekends have no forex data records), and is deemed sufficient for model training. The data is integrated into a table, and technical indices are calculated using ta, a technical analysis library for Python built on Pandas and Numpy.

Figure 6. EUR/USD exchange rate from 2010 to 2019.

Summing the inputs from technical analysis, fundamental analysis, and pure price data, a total of 1049 inputs are fed into the MLP. Within the hidden layers, ReLU activation is used, and a sigmoid activation function is used for the output layer. The output has a shape of 7×3, which represents the probability of the seven currency pairs and the three actions (buy, sell, do nothing).

Fig. 7 shows the accumulative profit from 2,000,000 steps in a single episode and its win rate (percentage of profitable trades within a moving average). An increasing spread value from 0.00001 to 0.00004 is applied, which the spread value starts from 0.00001 and increases by 0.00001 every 50,000 step. It can be seen that overall, the accumulative profit rises steadily. However, the win rate usually falls below the 50% line. How could a profitable trading strategy be possible? This is due to the fact that the average profit of a winning trade (=0.003736) is larger than the average loss of a losing trade (=0.003581). Thus, the overall result is a profitable trading strategy.

Figure 7. Accumulative profit and win rate from the training procedure of 2,000,000 steps.

Conclusion

In conclusion, a trading model for profitable forex trading is developed using reinforcement learning. The model can automatically adapt to dynamic environments to maximize its profits. Although for real conditions that have a larger spread, the model hasn’t achieved a stable and profitable result, the potential for optimizing is promising. In the future, I am planning to integrate this trading model with the automated forex trading system that I have made, and become a competitive player in this fascinating game of forex.

[Source code of RL model training section]

References

[1] R.S. Sutton, A.G. Barto, Reinforcement Learning: An Introduction, MIT Press2018.

Aroma Alarm Clock

As an old saying goes, “An hour in the morning is worth two in the evening”. What we encounter and conceive during this period is largely intertwined with our whole day. However, what is the most important thing that we encounter in the morning? It is about waking up. Nowadays, we wake up by the noisy, buzzing sound of an alarm. How would it be like if we could wake up by a scent, a fragrance, wiping away the unsatisfactory memories of the past? What can be changed to fulfill a better awakening?

An aroma alarm clock should fill in the answer. When I was a junior, I enrolled in a program called “NTU Creativity and Entrepreneurship Program” (NTUCEP). In this program, students with similar ideas form groups and learn about entrepreneurship by practice. After a couple days of brainstorming, our group came up with the concept about developing an alarm clock that can wake people up using scent packs. Apart from being awakened by a loud ringing sound of an alarm, fragrances (as we had proposed) have a milder, progressive effect for waking people up. We named our product “Sensio”, a Latin noun of the meaning “thought”.

Market Survey

In order to verify the market demand of this proposal, we devised a questionnaire for investigating the views and habits of people about waking up, which 1320 valid responses were collected. A score of 0 ~ 7 is used to quantify the “discomfort level” for waking up, in which a level of 0 ~ 3 is regarded as painless, and 4 ~ 7 as painful. Fig. 1 shows the discomfort level distribution of all the samples, and Fig. 2 shows the distributions by occupation. It can be seen that regardless of identity, the majority of people feel painful when waking up (students: 71%, office worker: 62%, other: 51%, overall: 68%). Moreover, we analyzed their habits of lying in bed after awakening. Fig. 3 shows the proportion and durations for lying in bed. Accordingly, over 44% of people lie in bed for over 15 minutes, and the average duration is 15.07 minutes. Theoretically, this means if a product could help lower one’s discomfort level of waking up, and reduce the duration of lying in bed after awakening, he could approximately increase 3.8 days/year of effective time. Not only having a favorable morning, but also enhancing his work efficiency.

Figure 1. Discomfort level distribution of waking up.

Figure 2. Number of people vs discomfort level by occupation.

Figure 3. Duration of lying in bed after awakening.

From the above analysis, we became confident about the large market demand for a better quality of getting up. For me, being the leading engineer role in our group of six meant turning ideas into reality. This had been a difficult period but a time of inspiration and excitement. I put into practice what I had learned within the past few years for developing a system, a hardware that can really be used as a product.

Sensio v0.1

I developed two preliminary versions of the hardware, with the 1st version using Arduino as the CPU (Sensio v0.1) (Fig. 4). In this version, an Arduino chip, a time clock module (DS1302), a shift register (74HC595N), relays, button switches, LED displays, a buzzer, a fan and a potentiometer are used for the alarm clock (Fig. 5). 10 different themes are written inside the system and a few simple features are integrated within (Fig. 6 details every mode of the alarm clock and the corresponding function of the buttons). This version serves as the proof-of-concept version for making an aroma alarm clock. Although being simple and somewhat rough, it demonstrated to my team and our professors a solid proof and feasibility of our objective.

[Source code for Sensio v0.1]

Figure 4. Photograph of Sensio v0.1.

Figure 5. Hardware architecture of Sensio v0.1.

Figure 6. Alarm modes and button function for Sensio v0.1.

During this period, we earned 40,000NTD in a fundraising event in our program, and won the Best Potential Award in the 14th NTU innovation competition. Fig. 7 shows the poster we had used in the competition. During that period, we had great enthusiasm and motivation, and dreamed that one day our product would become on-sale and we could establish a real listed company of our own. We entered the NTU garage, which is an incubator for startups that are closely linked to NTU.

Figure 7. Poster of product Sensio for the NTU innovation competition.

Sensio v0.2

The second version Sensio v0.2 was created during my second semester as a junior. This time, an IC chip (ATMEGA328P) is used as the CPU rather than Arduino in order to lower the cost of the system and decrease its volume. Moreover, a PCB layout for fixing electronic parts on a solid panel is used. Fig. 8 illustrates the structural model of Sensio v0.2, Fig. 9 shows a photograph of the integrated electric components and its corresponding PCB layout schematic, and Fig. 10 shows photographs of the laser-cut 3D outer case.

[Source code for Sensio v0.2]

Figure 8. Structural model of Sensio v0.2.

Figure 9. Electronic component-integrated PCB board of Sensio v0.2 and its corresponding layout schematic.

Figure 10. Outer case (blue) and scent pack holder (orange) of Sensio v0.2.

We were thrilled after these accomplishments, since never had a team of the same period achieved such a high level of completeness, especially when concerning the integrity of hardware and software. These remarkable things happen so quickly, so fleeting for us to catch up with. The truth is, we are still so far from success, since we were still students back then. Reality pulled us back, and we started asked ourselves: are we ready to give up some things and dedicate ourselves to entrepreneurship? Our dream product, the aroma alarm clock, had once seemed so close, yet had never seemed so far.

Although in the end, the functional aroma alarm clock wasn’t successfully made, what I’ve learned during this event is unprecedented compared with my past experience, not only techniques for hardware and software development, but also important philosophies for entrepreneurship. This project inspired me to a great extent, and enabled me for creating future hardware-related projects (e.g. Automated Microfluidic Controlling Platform, Real-time Impedance Detection Systems, Surface Plasmon Resonance Platform) in my own field of research.

Figure 11. Photograph of our team and professor.