Alex Blank
700 lines
52 KiB
TeX
700 lines
52 KiB
TeX
%! Author = alex
|
||
%! Date = 3/6/25
|
||
|
||
|
||
\section{Results}\label{sec:results}
|
||
|
||
We summarize the main findings from our modeling experiments,
|
||
beginning with overall model performance across architectures,
|
||
followed by a detailed analysis of the individual architectures' performances for irregular and regular cycles and the use cases
|
||
introduced in the last section.
|
||
These results will also be compared with the three baseline methods to evaluate their performance versus less sophisticated methods.
|
||
|
||
We summarize the main findings from our modeling experiments, beginning with overall model performance across architectures.
|
||
We then provide a detailed analysis of model performance for irregular and regular cycles, as well as for
|
||
the pregnancy and contraception use-cases described earlier.
|
||
Additionally, these results will be compared to those of the three baseline methods to evaluate the benefit of more advanced modeling approaches.
|
||
|
||
\subsection{Overall Model Performance Across Architectures}\label{subsec:overall_model_performance_across_architectures}
|
||
|
||
We evaluated multiple model architectures to compare their effectiveness in predicting the fertility-probability and ovulation-over targets.
|
||
Based on these results, we selected the best-performing configuration for each architecture for further analysis.
|
||
|
||
\subsubsection{Fertility-Probability Prediction}\label{subsec:fertility_probability_prediction}
|
||
|
||
This section examines model performance in predicting the probability of fertility,
|
||
focusing on the effects of input window length, input resolution, and key architecture parameters.
|
||
|
||
\paragraph{Impact of Input Window Length:}\label{subsubsec:fert_impact_of_historical_context}
|
||
Table~\ref{tab:fertility_results_by_window_length} summarizes the effect of varying the input sequence length on model performance for the fertility-probability target.
|
||
Across all architectures, no single window length consistently outperformed others across all metrics.
|
||
|
||
For the \textbf{LSTM}, the longest input (160~days) yielded the lowest overall MAE (0.0378) and the best
|
||
fertile-day performance (MAE~$=0.0834$, MSE~$=0.0102$), while shorter sequences tended to perform slightly worse
|
||
, particularly for fertile-day prediction.
|
||
Non-fertile day performance was best at 20~days (MSE~$=0.0013$).
|
||
|
||
In the \textbf{Transformer}, the optimal MAE for overall fertility occurred at 160~days (0.0411),
|
||
but the lowest fertile-day error was achieved at 80~days (MAE~$=0.0859$).
|
||
The best non-fertile-day performance was seen with 160~days (MSE~$=0.0016$).
|
||
|
||
For the \textbf{Convolution-LSTM}, the shortest windows generally underperformed, with the best overall MAE (0.0394) and MSE (0.0045) obtained at 40~days.
|
||
Fertile-day metrics were optimal at 40~days as well, while non-fertile-day performance peaked at 20~days (MSE~$=0.0013$).
|
||
|
||
The \textbf{Convolution-Transformer} achieved the global best MAE (0.0376) and MSE (0.0041) for overall fertility at 40~days,
|
||
indicating that intermediate historical context was most effective for this architecture.
|
||
Fertile-day performance was strongest at 20~days (MSE~$=0.0102$), while non-fertile-day predictions benefited from shorter inputs (10 or 40~days).
|
||
|
||
Overall, results suggest that the optimal input length is architecture-dependent, with intermediate windows (20--40~days)
|
||
frequently yielding competitive or best performance, while extremely long sequences (160~days) only benefited certain architectures such as the LSTM\@.
|
||
|
||
\begin{landscape}
|
||
\begin{table}
|
||
\small
|
||
\begin{tabularx}{\linewidth}{l*{6}{X}}
|
||
\toprule
|
||
\multirow{2}{*}{Input-Length in Days} & \multicolumn{3}{c}{MAE} & \multicolumn{3}{c}{MSE} \\
|
||
\cmidrule(r){2-4} \cmidrule(r){5-7}
|
||
& Fertility Overall & Fertile Days & Non-Fertile Days & Fertility Overall & Fertile Days & Non-Fertile Days \\
|
||
\midrule
|
||
\multicolumn{7}{c}{\textbf{LSTM}} \\ \midrule
|
||
10 & 0.0399 & 0.0857 & 0.0212 & 0.0045 & 0.0105 & 0.0021 \\
|
||
20 & 0.0421 & 0.0994 & 0.0180 & 0.0052 & 0.0144 & \underline{\textbf{0.0013}} \\
|
||
40 & 0.0406 & 0.0938 & 0.0184 & 0.0049 & 0.0128 & 0.0016 \\
|
||
80 & 0.0420 & 0.1006 & \underline{\textbf{0.0176}} & 0.0053 & 0.0147 & 0.0014 \\
|
||
160 & \underline{0.0378} & \underline{\textbf{0.0834}} & 0.0192 & \underline{0.0043} & \underline{\textbf{0.0102}} & 0.0019 \\
|
||
\midrule
|
||
\multicolumn{7}{c}{\textbf{Transformer}} \\ \midrule
|
||
10 & 0.0443 & 0.0868 & 0.0273 & 0.0046 & 0.0110 & 0.0021 \\
|
||
20 & 0.0472 & 0.0949 & 0.0274 & 0.0051 & 0.0133 & 0.0018 \\
|
||
40 & 0.0413 & 0.0861 & 0.0233 & \underline{0.0044} & 0.0111 & 0.0017 \\
|
||
80 & 0.0437 & \underline{0.0859} & 0.0269 & 0.0045 & \underline{0.0108} & 0.0021 \\
|
||
160 & \underline{0.0411} & 0.0882 & \underline{0.0218} & 0.0045 & 0.0116 & \underline{0.0016} \\
|
||
\midrule
|
||
\multicolumn{7}{c}{\textbf{Convolution-LSTM}} \\ \midrule
|
||
10 & 0.0435 & 0.0954 & 0.0224 & 0.0050 & 0.0133 & 0.0017 \\
|
||
20 & 0.0414 & 0.0996 & \underline{0.0179} & 0.0050 & 0.0145 & \underline{\textbf{0.0013}} \\
|
||
40 & \underline{0.0394} & \underline{0.0911} & 0.0184 & \underline{0.0045} & \underline{0.0122} & 0.0014 \\
|
||
80 & 0.0481 & 0.1005 & 0.0278 & 0.0054 & 0.0146 & 0.0019 \\
|
||
160 & 0.0505 & 0.1067 & 0.0290 & 0.0060 & 0.0166 & 0.0020 \\
|
||
\midrule
|
||
\multicolumn{7}{c}{\textbf{Convolution-Transformer}} \\ \midrule
|
||
10 & 0.0459 & 0.0962 & 0.0256 & 0.0049 & 0.0136 & \underline{0.0015} \\
|
||
20 & 0.0423 & \underline{0.0848} & 0.0257 & 0.0045 & \underline{\textbf{0.0102}} & 0.0022 \\
|
||
40 & \underline{\textbf{0.0376}} & 0.0854 & \underline{0.0186} & \underline{\textbf{0.0041}} & 0.0108 & \underline{0.0015} \\
|
||
80 & 0.0414 & 0.0932 & 0.0210 & 0.0048 & 0.0128 & 0.0018 \\
|
||
160 & 0.0523 & 0.1018 & 0.0337 & 0.0059 & 0.0148 & 0.0026 \\
|
||
\bottomrule
|
||
\end{tabularx}
|
||
\caption{Evaluation Metrics for the Fertility-Probability Target across Different Model Architectures and Input Lengths on a fixed Input Resolution of 12 Values per Day.
|
||
\underline{Underlined} values represent the best value for each metric within a model.
|
||
\textbf{\underline{Bold + Underlined}} values represent the global best values across all models for a given metric.}
|
||
\label{tab:fertility_results_by_window_length}
|
||
\end{table}
|
||
\end{landscape}
|
||
|
||
\paragraph{Impact of Input Resolution:}\label{subsubsec:fert_impact_of_input_resolution}
|
||
|
||
Table~\ref{tab:fertility_results_by_window_resolution} shows the effect of varying the input resolution (values per day)
|
||
on model performance for the fertility-probability target.
|
||
Note, that the convolutional models are not included here, as they have their own learned input representation via convolution.
|
||
No single resolution consistently outperformed others across all metrics, and optimal settings varied by architecture.
|
||
|
||
For the \textbf{LSTM}, the lowest overall MAE (0.0402) was obtained at 48~values/day, while the best overall MSE (0.0046) occurred at 4~values/day.
|
||
Fertile-day performance was optimal at 288~values/day (MAE~$=0.0872$, MSE~$=0.0110$), and non-fertile-day metrics were best at
|
||
12~values/day (MAE~$=0.0180$, MSE~$=0.0013$), both of which represent the global best values for these categories.
|
||
|
||
In the \textbf{Transformer}, the lowest overall MAE (0.0447) occurred at 48~values/day, while the best overall MSE (0.0048) was shared between 48~and 288~values/day.
|
||
Fertile-day performance peaked at 4~values/day (MAE~$=0.0891$, MSE~$=0.0115$),
|
||
whereas non-fertile-day metrics were best at 48~values/day (MAE~$=0.0254$) and 72~values/day (MSE~$=0.0014$).
|
||
|
||
Overall, the results indicate that intermediate input resolutions (4--48~values/day) often yielded the best overall performance,
|
||
while extreme resolutions (1 or 288~values/day) only benefited specific metrics such as fertile-day prediction for the LSTM\@.
|
||
|
||
\begin{landscape}
|
||
\begin{table}
|
||
\small
|
||
\begin{tabularx}{\linewidth}{l*{6}{X}}
|
||
\toprule
|
||
\multirow{2}{*}{Values Per Day} & \multicolumn{3}{c}{MAE} & \multicolumn{3}{c}{MSE} \\
|
||
\cmidrule(r){2-4} \cmidrule(r){5-7}
|
||
& Fertility Overall & Fertile Days & Non-Fertile Days & Fertility Overall & Fertile Days & Non-Fertile Days \\
|
||
\midrule
|
||
\multicolumn{7}{c}{\textbf{LSTM}} \\
|
||
\midrule
|
||
1 & 0.0471 & 0.0987 & 0.0220 & 0.0062 & 0.0145 & 0.0022 \\
|
||
2 & 0.0457 & 0.0962 & 0.0231 & 0.0052 & 0.0133 & 0.0016 \\
|
||
4 & 0.0421 & 0.0898 & 0.0223 & \underline{\textbf{0.0046}}& 0.0116 & 0.0018 \\
|
||
12 & 0.0421 & 0.0994 & \underline{\textbf{0.0180}}& 0.0052 & 0.0144 & \underline{\textbf{0.0013}}\\
|
||
24 & 0.0419 & 0.0949 & 0.0201 & 0.0050 & 0.0132 & 0.0017 \\
|
||
48 & \underline{\textbf{0.0402}}& 0.0929 & 0.0182 & 0.0049 & 0.0126 & 0.0018 \\
|
||
72 & 0.0433 & 0.0972 & 0.0216 & 0.0052 & 0.0137 & 0.0018 \\
|
||
288 & 0.0410 & \underline{\textbf{0.0872}}& 0.0223 & 0.0049 & \underline{\textbf{0.0110}}& 0.0025 \\
|
||
\midrule
|
||
\multicolumn{7}{c}{\textbf{Transformer}} \\
|
||
\midrule
|
||
1 & 0.0541 & 0.1015 & 0.0313 & 0.0063 & 0.0146 & 0.0023 \\
|
||
2 & 0.0468 & 0.0946 & 0.0259 & 0.0054 & 0.0128 & 0.0021 \\
|
||
4 & 0.0456 & \underline{0.0891}& 0.0277 & 0.0050 & \underline{0.0115}& 0.0023 \\
|
||
12 & 0.0472 & 0.0949 & 0.0274 & 0.0051 & 0.0133 & 0.0018 \\
|
||
24 & 0.0502 & 0.0960 & 0.0323 & 0.0052 & 0.0133 & 0.0021 \\
|
||
48 & 0.0447 & 0.0912 & \underline{0.0254}& \underline{0.0048}& 0.0122 & 0.0017 \\
|
||
72 & 0.0468 & 0.0975 & 0.0257 & 0.0051 & 0.0140 & \underline{0.0014} \\
|
||
288 & \underline{0.0449}& 0.0937 & 0.0257 & \underline{0.0048}& 0.0128 & 0.0017 \\
|
||
\bottomrule
|
||
\end{tabularx}
|
||
\caption{Evaluation Metrics for the Fertility-Probability Target across Different Model Architectures and Input Resolutions on a fixed Input-Window-Length of 20 Days.
|
||
\underline{Underlined} values represent the best value for each metric within a model.
|
||
\textbf{\underline{Bold + Underlined}} values represent the global best values across all models for a given metric.}
|
||
\label{tab:fertility_results_by_window_resolution}
|
||
\end{table}
|
||
\end{landscape}
|
||
|
||
\paragraph{Impact of Model Parameters:}\label{subsubsec:fert_impaoct_of_model_parameters}
|
||
|
||
Tables~\ref{tab:fertility_results_by_model_parameters_lstm}--\ref{tab:fertility_results_by_model_parameters_conv_transformer} report
|
||
the results of the model parameter search across all architectures.
|
||
Each table shows the effect of varying hidden layer size and number of LSTM layers (for recurrent models),
|
||
or embedding size, number of encoder layers, and attention heads (for Transformer-based models).
|
||
|
||
For the \textbf{LSTM}, performance improved with increasing hidden layer size,
|
||
reaching the best overall values at 512 units with four layers (MAE~$=0.0399$, MSE~$=0.0047$). The lowest fertile-day errors were also
|
||
observed in this configuration, while non-fertile-day performance peaked at 256 units (MAE~$=0.0175$, MSE~$=0.0011$).
|
||
|
||
In the \textbf{Transformer}, larger embeddings and deeper networks generally improved performance.
|
||
The best overall MAE (0.0395) was achieved with a 512-dimensional embedding, eight encoder layers, and eight attention heads.
|
||
The lowest fertile-day errors occurred with a 512-dimensional embedding and four layers (MAE~$=0.0831$,
|
||
MSE~$=0.0103$), whereas non-fertile-day performance was strongest at 512~×~8 (MAE~$=0.0159$, MSE~$=0.0011$).
|
||
|
||
For the \textbf{Convolution-LSTM}, the best overall configuration was 256 hidden units with four layers,
|
||
yielding the lowest overall MSE (0.0042) and fertile-day MSE (0.0100).
|
||
Increasing to 512 units slightly reduced overall MAE (0.0380) and non-fertile-day MAE (0.0184).
|
||
|
||
In the \textbf{Convolution-Transformer}, the optimal configuration used a 512-dimensional embedding with four encoder layers and four attention heads,
|
||
achieving the best overall MAE (0.0377) and non-fertile-day performance (MAE~$=0.0177$, MSE~$=0.0013$).
|
||
Fertile-day prediction was strongest with eight encoder layers (MAE~$=0.0842$, MSE~$=0.0107$).
|
||
|
||
Taken together, these results indicate that larger model capacities generally improved performance across all architectures,
|
||
with the best configurations typically found at the higher end of the tested parameter ranges.
|
||
|
||
\begin{landscape}
|
||
\begin{table}
|
||
\scriptsize
|
||
\begin{tabularx}{\linewidth}{l*{8}{X}}
|
||
\midrule
|
||
\multicolumn{8}{c}{\textbf{LSTM}} \\
|
||
\toprule
|
||
\multirow{2}{*}{Hidden Layer Size} &
|
||
\multirow{2}{*}{\# LSTM Layers} &
|
||
\multicolumn{3}{c}{MAE} &
|
||
\multicolumn{3}{c}{MSE} \\
|
||
\cmidrule(lr){3-5} \cmidrule(lr){6-8}
|
||
& & Fert Overall & Fert Days & Non-Fert Days & Fert Overall & Fert Days & Non-Fert Days \\
|
||
\midrule
|
||
16 & 1 & 0.0485 & 0.1050 & 0.0252 & 0.0059 & 0.0161 & 0.0016 \\
|
||
32 & 1 & 0.0477 & 0.1029 & 0.0248 & 0.0057 & 0.0155 & 0.0016 \\
|
||
32 & 2 & 0.0458 & 0.0922 & 0.0269 & 0.0052 & \textbf{0.0122} & 0.0024 \\
|
||
64 & 2 & 0.0447 & 0.0925 & 0.0250 & 0.0051 & 0.0123 & 0.0021 \\
|
||
128 & 2 & 0.0424 & 0.0951 & 0.0201 & 0.0051 & 0.0131 & 0.0018 \\
|
||
128 & 4 & 0.0427 & 0.0989 & 0.0191 & 0.0053 & 0.0145 & 0.0014 \\
|
||
256 & 4 & 0.0433 & 0.1050 & \textbf{0.0175} & 0.0055 & 0.0161 & \textbf{0.0011} \\
|
||
512 & 4 & \textbf{0.0399} & \textbf{0.0911} & 0.0189 & \textbf{0.0047} & \textbf{0.0122} & 0.0016 \\
|
||
\bottomrule
|
||
\end{tabularx}
|
||
\caption{Evaluation Metrics for the Fertility-Probability Target across Different Model Parameters for the LSTM model with
|
||
a fixed input window length of 160 days and an input resolution of 12 values per day.
|
||
\textbf{Bold} values represent the best value for each metric within a model.}
|
||
\label{tab:fertility_results_by_model_parameters_lstm}
|
||
\end{table}
|
||
\begin{table}
|
||
\scriptsize
|
||
\begin{tabularx}{\linewidth}{l*{8}{X}}
|
||
\midrule
|
||
\multicolumn{9}{c}{\textbf{Transformer}} \\
|
||
\toprule
|
||
\multirow{2}{*}{Size of Embedding} &
|
||
\multirow{2}{*}{\# Encoder Layers} &
|
||
\multirow{2}{*}{\# Attention Heads} &
|
||
\multicolumn{3}{c}{MAE} &
|
||
\multicolumn{3}{c}{MSE} \\
|
||
\cmidrule(lr){4-6} \cmidrule(lr){7-9}
|
||
& & & Fert Overall & Fert Days & Non-Fert Days & Fert Overall & Fert Days & Non-Fert Days \\
|
||
\midrule
|
||
16 & 1 & 1 & 0.0708 & 0.1262 & 0.0475 & 0.0084 & 0.0227 & 0.0024 \\
|
||
32 & 1 & 1 & 0.0599 & 0.1163 & 0.0378 & 0.0070 & 0.0196 & 0.0019 \\
|
||
64 & 1 & 1 & 0.0601 & 0.1178 & 0.0373 & 0.0071 & 0.0200 & 0.0020 \\
|
||
64 & 2 & 2 & 0.0477 & 0.0908 & 0.0301 & 0.0049 & 0.0121 & 0.0020 \\
|
||
128 & 2 & 2 & 0.0449 & 0.0904 & 0.0263 & 0.0047 & 0.0119 & 0.0017 \\
|
||
128 & 4 & 4 & 0.0461 & 0.0979 & 0.0247 & 0.0050 & 0.0143 & 0.0013 \\
|
||
256 & 4 & 4 & 0.0475 & 0.0919 & 0.0293 & 0.0048 & 0.0125 & 0.0017 \\
|
||
512 & 4 & 4 & 0.0403 & \textbf{0.0831} & 0.0229 & \textbf{0.0043} & \textbf{0.0103} & 0.0018 \\
|
||
512 & 8 & 8 & \textbf{0.0395} & 0.0967 & \textbf{0.0159} & 0.0048 & 0.0139 & \textbf{0.0011} \\
|
||
\bottomrule
|
||
\end{tabularx}
|
||
\caption{Evaluation Metrics for the Fertility-Probability Target across Different Model Parameters for the Transformer model with
|
||
a fixed input window length of 160 days and an input resolution of 12 values per day.
|
||
\textbf{Bold} values represent the best value for each metric within a model.}
|
||
\label{tab:fertility_results_by_model_parameters_transformer}
|
||
\end{table}
|
||
|
||
\begin{table}
|
||
\scriptsize
|
||
\begin{tabularx}{\linewidth}{l*{8}{X}}
|
||
\midrule
|
||
\multicolumn{8}{c}{\textbf{Convolutional-LSTM}} \\
|
||
\toprule
|
||
\multirow{2}{*}{Hidden Layer Size} &
|
||
\multirow{2}{*}{\# LSTM Layers} &
|
||
\multicolumn{3}{c}{MAE} &
|
||
\multicolumn{3}{c}{MSE} \\
|
||
\cmidrule(lr){3-5} \cmidrule(lr){6-8}
|
||
& & Fert Overall & Fert Days & Non-Fert Days & Fert Overall & Fert Days & Non-Fert Days \\
|
||
\midrule
|
||
16 & 1 & 0.0487 & 0.1029 & 0.0264 & 0.0056 & 0.0154 & \textbf{0.0016} \\
|
||
32 & 1 & 0.0461 & 0.0956 & 0.0266 & 0.0050 & 0.0133 & 0.0017 \\
|
||
32 & 2 & 0.0448 & 0.0975 & 0.0232 & 0.0051 & 0.0139 & \textbf{0.0016} \\
|
||
64 & 2 & 0.0432 & 0.0921 & 0.0241 & 0.0048 & 0.0123 & 0.0019 \\
|
||
128 & 2 & 0.0400 & 0.0895 & 0.0204 & 0.0045 & 0.0119 & \textbf{0.0016} \\
|
||
128 & 4 & 0.0398 & 0.0871 & 0.0204 & 0.0044 & 0.0113 & \textbf{0.0016} \\
|
||
256 & 4 & 0.0389 & \textbf{0.0826} & 0.0209 & \textbf{0.0042} & \textbf{0.0100} & 0.0018 \\
|
||
512 & 4 & \textbf{0.0380} & 0.0869 & \textbf{0.0184} & 0.0044 & 0.0112 & 0.0017 \\
|
||
\bottomrule
|
||
\end{tabularx}
|
||
\caption{Evaluation Metrics for the Fertility-Probability Target across Different Model Parameters for the convolutional LSTM model with
|
||
a fixed input window length of 40 days.
|
||
\textbf{Bold} values represent the best value for each metric within a model.}
|
||
\label{tab:fertility_results_by_model_parameters_conv_lstm}
|
||
\end{table}
|
||
\begin{table}
|
||
\scriptsize
|
||
\begin{tabularx}{\linewidth}{l*{8}{X}}
|
||
\midrule
|
||
\multicolumn{8}{c}{\textbf{Convolutional-Transformer}} \\
|
||
\toprule
|
||
\multirow{2}{*}{Size of Embedding} &
|
||
\multirow{2}{*}{\# Encoder Layers} &
|
||
\multirow{2}{*}{\# Attention Heads} &
|
||
\multicolumn{3}{c}{MAE} &
|
||
\multicolumn{3}{c}{MSE} \\
|
||
\cmidrule(lr){4-6} \cmidrule(lr){7-9}
|
||
& & & Fert Overall & Fert Days & Non-Fert Days & Fert Overall & Fert Days & Non-Fert Days \\
|
||
\midrule
|
||
16 & 1 & 1 & 0.0574 & 0.1005 & 0.0427 & 0.0061 & 0.0146 & 0.0032 \\
|
||
32 & 1 & 1 & 0.0462 & 0.0963 & 0.0266 & 0.0051 & 0.0135 & 0.0018 \\
|
||
64 & 1 & 1 & 0.0546 & 0.1006 & 0.0379 & 0.0059 & 0.0145 & 0.0027 \\
|
||
64 & 2 & 2 & 0.0427 & 0.0903 & 0.0237 & 0.0046 & 0.0118 & 0.0018 \\
|
||
128 & 2 & 2 & 0.0406 & 0.0868 & 0.0225 & 0.0044 & 0.0113 & 0.0017 \\
|
||
128 & 4 & 4 & 0.0411 & 0.0886 & 0.0230 & 0.0044 & 0.0116 & 0.0017 \\
|
||
256 & 4 & 4 & 0.0420 & 0.0866 & 0.0240 & \textbf{0.0043} & 0.0113 & 0.0016 \\
|
||
512 & 4 & 4 & \textbf{0.0377} & 0.0878 & \textbf{0.0177} & \textbf{0.0043} & 0.0117 & \textbf{0.0013} \\
|
||
512 & 8 & 8 & 0.0399 & \textbf{0.0842} & 0.0224 & 0.0044 & \textbf{0.0107} & 0.0019 \\
|
||
\bottomrule
|
||
\end{tabularx}
|
||
\caption{Evaluation Metrics for the Fertility-Probability Target across Different Model Parameters for the convolutional Transformer model with
|
||
a fixed input window length of 40 days.
|
||
\textbf{Bold} values represent the best value for each metric within a model.}
|
||
\label{tab:fertility_results_by_model_parameters_conv_transformer}
|
||
\end{table}
|
||
\end{landscape}
|
||
|
||
\subsubsection{Ovulation-Over Prediction}\label{subsubsec:ov_over_prediction}
|
||
|
||
This section examines model performance in predicting the indicator for a passed ovulation, with attention to the influence of input window length,
|
||
the input resolution and the individual model architecture parameters.
|
||
|
||
\paragraph{Impact of Input Window Length:}\label{subsubsec:ov_over_impact_of_historical_context}
|
||
|
||
Table~\ref{tab:ov_over_results_by_window_length} reports the impact of input sequence length on performance for the ovulation-over target.
|
||
The optimal length varied across architectures and prediction phases.
|
||
|
||
For the \textbf{LSTM}, the best overall MAE (0.1039) and MSE (0.0557) were obtained with 160~days,
|
||
which also gave the lowest after-ovulation errors (MAE~$=0.0973$, MSE~$=0.0580$). However, the best before-ovulation
|
||
performance occurred at 20~days (MAE~$=0.0745$, MSE~$=0.0212$), which were the global bests for this phase.
|
||
|
||
In the \textbf{Transformer}, 40~days yielded the lowest overall MAE (0.1017) and MSE (0.0533),
|
||
as well as the lowest after-ovulation errors (MAE~$=0.0883$, MSE~$=0.0520$), all of which were global bests.
|
||
The best before-ovulation results were achieved at 80~days (MAE~$=0.0897$) and 20~days (MSE~$=0.0255$).
|
||
|
||
For the \textbf{Convolution-LSTM}, the shortest effective length was 40~days,
|
||
which achieved the lowest overall MAE (0.1455) and before-ovulation MAE (0.1089).
|
||
The best overall MSE (0.0699) and before-ovulation MSE (0.0389) were observed at 20~days.
|
||
After-ovulation errors were smallest at 20~days (MSE~$=0.0833$) and 40~days (MAE~$=0.1561$).
|
||
|
||
In the \textbf{Convolution-Transformer}, the best overall MAE (0.1448) and MSE (0.0709) occurred at 40~days,
|
||
which also minimized after-ovulation MAE (0.1514) and MSE (0.0820).
|
||
The best before-ovulation performance came from 160~days for MAE (0.1051) and 10~days for MSE (0.0317).
|
||
|
||
Overall, intermediate input lengths (20--40~days) were often optimal, particularly for before-ovulation prediction,
|
||
while longer sequences (160~days) occasionally improved after-ovulation accuracy.
|
||
|
||
\begin{landscape}
|
||
\begin{table}
|
||
\small
|
||
\begin{tabularx}{\linewidth}{l*{6}{X}}
|
||
\toprule
|
||
\multirow{2}{*}{Input-Length in Days} & \multicolumn{3}{c}{MAE} & \multicolumn{3}{c}{MSE} \\
|
||
\cmidrule(r){2-4} \cmidrule(r){5-7}
|
||
& OV-Over Overall & OV-Over Before OV & OV-Over After OV & OV-Over Overall & OV-Over Before OV & OV-Over After OV \\
|
||
\midrule
|
||
\multicolumn{7}{c}{\textbf{LSTM}} \\
|
||
\midrule
|
||
10 & 0.1218 & 0.1044 & 0.1223 & 0.0612 & 0.0289 & 0.0711 \\
|
||
20 & 0.1153 & \underline{\textbf{0.0745}} & 0.1312 & 0.0641 & \underline{\textbf{0.0212}} & 0.0822 \\
|
||
40 & 0.1066 & 0.0820 & 0.1128 & 0.0616 & 0.0281 & 0.0740 \\
|
||
80 & 0.1173 & 0.0842 & 0.1291 & 0.0647 & 0.0263 & 0.0801 \\
|
||
160 & \underline{0.1039} & 0.1139 & \underline{0.0973} & \underline{0.0557} & 0.0462 & \underline{0.0580} \\
|
||
\midrule
|
||
\multicolumn{7}{c}{\textbf{Transformer}} \\
|
||
\midrule
|
||
10 & 0.1137 & 0.1006 & 0.1186 & 0.0618 & 0.0336 & 0.0740 \\
|
||
20 & 0.1204 & 0.0771 & 0.1366 & 0.0690 & \underline{0.0255} & 0.0864 \\
|
||
40 & \underline{\textbf{0.1017}} & 0.1409 & \underline{\textbf{0.0883}} & \underline{\textbf{0.0533}} & 0.0621 & \underline{\textbf{0.0520}} \\
|
||
80 & 0.1138 & \underline{0.0897} & 0.1234 & 0.0654 & 0.0356 & 0.0788 \\
|
||
160 & 0.1076 & 0.0959 & 0.1141 & 0.0606 & 0.0379 & 0.0714 \\
|
||
\midrule
|
||
\multicolumn{7}{c}{\textbf{Convolution-LSTM}} \\
|
||
\midrule
|
||
10 & 0.1847 & 0.1878 & 0.1840 & 0.0819 & 0.0581 & 0.0949 \\
|
||
20 & 0.1493 & 0.1274 & 0.1562 & \underline{0.0699} & \underline{0.0389} & \underline{0.0833} \\
|
||
40 & \underline{0.1455} & \underline{0.1089} & \underline{0.1561} & 0.0722 & 0.0358 & 0.0852 \\
|
||
80 & 0.2327 & 0.1796 & 0.2507 & 0.1168 & 0.0686 & 0.1327 \\
|
||
160 & 0.2518 & 0.2040 & 0.2757 & 0.1293 & 0.0878 & 0.1503 \\
|
||
\midrule
|
||
\multicolumn{7}{c}{\textbf{Convolution-Transformer}} \\
|
||
\midrule
|
||
10 & 0.1472 & 0.1053 & 0.1651 & 0.0768 & 0.0317 & 0.0966 \\
|
||
20 & 0.1530 & 0.1307 & 0.1637 & 0.0745 & 0.0443 & 0.0886 \\
|
||
40 & \underline{0.1448} & 0.1228 & \underline{0.1514} & \underline{0.0709} & 0.0435 & \underline{0.0820} \\
|
||
80 & 0.1685 & 0.1164 & 0.1889 & 0.0865 & 0.0345 & 0.1089 \\
|
||
160 & 0.2440 & \underline{0.1051} & 0.3117 & 0.1403 & \underline{0.0286} & 0.1946 \\
|
||
\bottomrule
|
||
|
||
\end{tabularx}
|
||
\caption{Evaluation Metrics for the Ovulation-Over Target across Different Model Architectures and Input Lengths on a fixed Input Resolution of 12 Values per Day.
|
||
\underline{Underlined} values represent the best value for each metric within a model.
|
||
\textbf{\underline{Bold + Underlined}} values represent the global best values across all models for a given metric.}
|
||
\label{tab:ov_over_results_by_window_length}
|
||
\end{table}
|
||
\end{landscape}
|
||
|
||
\paragraph{Impact of Input Resolution:}\label{subsubsec:ov_over_impact_of_input_resolution}
|
||
|
||
Table~\ref{tab:ov_over_results_by_resolution} presents the effect of varying input resolution (values per day)
|
||
on performance for the ovulation-over target.
|
||
Note, that the convolutional models are not included here, as they have their own learned input representation via convolution.
|
||
The best-performing resolution differed across architectures and prediction phases.
|
||
|
||
For the \textbf{LSTM}, the lowest overall MAE (0.1153) was achieved at 12~values/day,
|
||
which also produced the global best before-ovulation results (MAE~$=0.0745$, MSE~$=0.0212$).
|
||
The best overall MSE (0.0633) was observed at 24~values/day.
|
||
After-ovulation performance was strongest at 48~values/day (MAE~$=0.0973$, MSE~$=0.0550$), the latter being a global best.
|
||
|
||
In the \textbf{Transformer}, the optimal overall MAE (0.1041) and MSE (0.0585) were both achieved at 72~values/day, which also yielded the
|
||
lowest after-ovulation MAE (0.0947), all of which were global bests.
|
||
The best before-ovulation MAE (0.0771) and MSE (0.0255) were found at 12~values/day.
|
||
The lowest after-ovulation MSE (0.0578) occurred at 288~values/day.
|
||
|
||
Overall, intermediate input resolutions (12--72~values/day) tended to perform best for ovulation-over prediction,
|
||
with 12~values/day favouring before-ovulation performance and 48--72~values/day improving after-ovulation accuracy.
|
||
|
||
\begin{landscape}
|
||
\begin{table}
|
||
\small
|
||
\begin{tabularx}{\linewidth}{l*{6}{X}}
|
||
\toprule
|
||
\multirow{2}{*}{Values Per Day} & \multicolumn{3}{c}{MAE} & \multicolumn{3}{c}{MSE} \\
|
||
\cmidrule(r){2-4} \cmidrule(r){5-7}
|
||
& OV-Over Overall & OV-Over Before OV & OV-Over After OV & OV-Over Overall & OV-Over Before OV & OV-Over After OV \\
|
||
\midrule
|
||
\multicolumn{7}{c}{\textbf{LSTM}} \\
|
||
\midrule
|
||
1 & 0.1691 & 0.1280 & 0.1819 & 0.0914 & 0.0411 & 0.1099 \\
|
||
2 & 0.1431 & 0.1239 & 0.1471 & 0.0747 & 0.0442 & 0.0849 \\
|
||
4 & 0.1344 & 0.1130 & 0.1417 & 0.0680 & 0.0371 & 0.0807 \\
|
||
12 & \underline{0.1153} & \underline{\textbf{0.0745}} & 0.1312 & 0.0641 & \underline{\textbf{0.0212}} & 0.0822 \\
|
||
24 & 0.1192 & 0.0975 & 0.1240 & \underline{0.0633} & 0.0288 & 0.0755 \\
|
||
48 & 0.1463 & 0.2493 & \underline{0.0973} & 0.0768 & 0.1203 & \underline{\textbf{0.0550}} \\
|
||
72 & 0.1587 & 0.1113 & 0.1844 & 0.0857 & 0.0341 & 0.1126 \\
|
||
288 & 0.1353 & 0.1801 & 0.1169 & 0.0726 & 0.0799 & 0.0713 \\
|
||
\midrule
|
||
\multicolumn{7}{c}{\textbf{Transformer}} \\
|
||
\midrule
|
||
1 & 0.1687 & 0.1582 & 0.1709 & 0.0883 & 0.0578 & 0.0996 \\
|
||
2 & 0.1469 & 0.0895 & 0.1676 & 0.0823 & 0.0260 & 0.1044 \\
|
||
4 & 0.1282 & 0.0999 & 0.1393 & 0.0704 & 0.0329 & 0.0862 \\
|
||
12 & 0.1204 & \underline{0.0771}& 0.1366 & 0.0690 & \underline{0.0255}& 0.0864 \\
|
||
24 & 0.1952 & 0.1649 & 0.2102 & 0.0904 & 0.0638 & 0.1040 \\
|
||
48 & 0.1137 & 0.1159 & 0.1073 & 0.0610 & 0.0480 & 0.0629 \\
|
||
72 & \underline{\textbf{0.1041}}& 0.1180 & \underline{\textbf{0.0947}}& \underline{\textbf{0.0585}}& 0.0538 & 0.0581 \\
|
||
288 & 0.1319 & 0.1879 & 0.1097 & 0.0617 & 0.0769 & \underline{0.0578}\\
|
||
\bottomrule
|
||
\end{tabularx}
|
||
\caption{Evaluation Metrics for the Ovulation-Over Target across Different Model Architectures and Input Resolutions on a fixed Input-Window-Length of 20 Days.
|
||
\underline{Underlined} values represent the best value for each metric within a model.
|
||
\textbf{\underline{Bold + Underlined}} values represent the global best values across all models for a given metric.}
|
||
\label{tab:ov_over_results_by_resolution}
|
||
\end{table}
|
||
\end{landscape}
|
||
|
||
\paragraph{Impact of Model Parameters:}\label{subsubsec:ov_over_impaoct_of_model_parameters}
|
||
|
||
Tables~\ref{tab:ov_over_results_by_model_parameters_lstm}--\ref{tab:ov_over_results_by_model_parameters_conv_transformer}
|
||
show the results of the parameter exploration for the ovulation-over target.
|
||
Each table reports the effect of varying hidden layer size and number of LSTM layers (for recurrent models),
|
||
or embedding size, number of encoder layers, and number of attention heads (for Transformer-based models).
|
||
|
||
For the \textbf{LSTM}, performance improved with increasing hidden layer size,
|
||
with the best overall MAE (0.1120) and MSE (0.0616) obtained at 512 units with four layers.
|
||
This configuration also yielded the lowest after-ovulation errors (MAE~$=0.0983$, MSE~$=0.0613$).
|
||
The lowest before- ovulation errors were observed at 128 units with two layers (MAE~$=0.0861$, MSE~$=0.0274$).
|
||
|
||
In the \textbf{Transformer}, smaller configurations performed poorly, while larger ones markedly improved results.
|
||
The best overall MAE (0.1114) was achieved with a 128-dimensional embedding and four encoder layers,
|
||
whereas the best overall MSE (0.0543) occurred with a 512-dimensional embedding and eight encoder layers.
|
||
The lowest before-ovulation errors were found at 64 dimensions with two layers (MAE~$=0.0810$, MSE~$=0.0302$),
|
||
while after-ovulation performance was best at 512 dimensions with eight layers (MAE~$=0.0753$, MSE~$=0.0410$).
|
||
|
||
For the \textbf{Convolution-LSTM}, the best overall configuration used 256 hidden units with four layers,
|
||
reaching the lowest overall MAE (0.1424) and MSE (0.0687).
|
||
This configuration also minimized after-ovulation errors (MAE~$=0.1369$, MSE~$=0.0715$).
|
||
Before-ovulation performance was strongest with 128 units and two layers (MAE~$=0.1138$, MSE~$=0.0357$).
|
||
|
||
In the \textbf{Convolution-Transformer}, the best overall MAE (0.1495) and MSE (0.0703)
|
||
were achieved with a 512-dimensional embedding and four encoder layers.
|
||
This configuration also gave the lowest after-ovulation errors (MAE~$=0.1551$, MSE~$=0.0814$).
|
||
Before-ovulation errors were lowest at 512 dimensions with eight layers (MAE~$=0.1025$, MSE~$=0.0339$).
|
||
|
||
Overall, the results show that larger configurations generally improved performance across architectures for the ovulation-over target,
|
||
with the best outcomes typically found at the higher-capacity settings.
|
||
|
||
\begin{landscape}
|
||
\begin{table}
|
||
\scriptsize
|
||
\begin{tabularx}{\linewidth}{l*{8}{X}}
|
||
\midrule
|
||
\multicolumn{8}{c}{\textbf{LSTM}} \\
|
||
\toprule
|
||
\multirow{2}{*}{Hidden Layer Size} &
|
||
\multirow{2}{*}{\# LSTM Layers} &
|
||
\multicolumn{3}{c}{MAE} &
|
||
\multicolumn{3}{c}{MSE} \\
|
||
\cmidrule(lr){3-5} \cmidrule(lr){6-8}
|
||
& & Fert Overall & Fert Days & Non-Fert Days & Fert Overall & Fert Days & Non-Fert Days \\
|
||
\midrule
|
||
16 & 1 & 0.2253 & 0.1353 & 0.2735 & 0.1196 & 0.0397 & 0.1621 \\
|
||
32 & 1 & 0.1746 & 0.1255 & 0.2005 & 0.0860 & 0.0404 & 0.1109 \\
|
||
32 & 2 & 0.1349 & 0.1240 & 0.1399 & 0.0674 & 0.0396 & 0.0786 \\
|
||
64 & 2 & 0.1270 & 0.0943 & 0.1365 & 0.0645 & 0.0292 & 0.0772 \\
|
||
128 & 2 & 0.1151 & \textbf{0.0861} & 0.1243 & 0.0626 & \textbf{0.0274} & 0.0755 \\
|
||
128 & 4 & 0.1326 & 0.1064 & 0.1407 & 0.0658 & 0.0291 & 0.0811 \\
|
||
256 & 4 & 0.1353 & 0.1111 & 0.1407 & 0.0662 & 0.0360 & 0.0765 \\
|
||
512 & 4 & \textbf{0.1120} & 0.1358 & \textbf{0.0983} & \textbf{0.0616} & 0.0603 & \textbf{0.0613} \\
|
||
\bottomrule
|
||
\end{tabularx}
|
||
\caption{Evaluation Metrics for the Ovulation-Over Target across Different Model Parameters for the LSTM model with
|
||
a fixed input window length of 160 days and an input resolution of 12 values per day.
|
||
\textbf{Bold} values represent the best value for each metric within a model.}
|
||
\label{tab:ov_over_results_by_model_parameters_lstm}
|
||
\end{table}
|
||
\begin{table}
|
||
\scriptsize
|
||
\begin{tabularx}{\linewidth}{l*{8}{X}}
|
||
\midrule
|
||
\multicolumn{9}{c}{\textbf{Transformer}} \\
|
||
\toprule
|
||
\multirow{2}{*}{Size of Embedding} &
|
||
\multirow{2}{*}{\# Encoder Layers} &
|
||
\multirow{2}{*}{\# Attention Heads} &
|
||
\multicolumn{3}{c}{MAE} &
|
||
\multicolumn{3}{c}{MSE} \\
|
||
\cmidrule(lr){4-6} \cmidrule(lr){7-9}
|
||
& & & Fert Overall & Fert Days & Non-Fert Days & Fert Overall & Fert Days & Non-Fert Days \\
|
||
\midrule
|
||
16 & 1 & 1 & 0.4792 & 0.5280 & 0.4675 & 0.2336 & 0.2820 & 0.2220 \\
|
||
32 & 1 & 1 & 0.4406 & 0.4572 & 0.4417 & 0.2046 & 0.2179 & 0.2068 \\
|
||
64 & 1 & 1 & 0.3614 & 0.3411 & 0.3796 & 0.1696 & 0.1604 & 0.1800 \\
|
||
64 & 2 & 2 & 0.1532 & \textbf{0.0810} & 0.1913 & 0.0904 & 0.0302 & 0.1214 \\
|
||
128 & 2 & 2 & 0.1476 & 0.0979 & 0.1739 & 0.0816 & 0.0332 & 0.1064 \\
|
||
128 & 4 & 4 & \textbf{0.1114} & 0.0887 & 0.1236 & 0.0668 & 0.0368 & 0.0821 \\
|
||
256 & 4 & 4 & 0.1310 & 0.0920 & 0.1433 & 0.0722 & \textbf{0.0293} & 0.0876 \\
|
||
512 & 4 & 4 & 0.1222 & 0.1047 & 0.1261 & 0.0610 & 0.0321 & 0.0708 \\
|
||
512 & 8 & 8 & 0.1126 & 0.1965 & \textbf{0.0753} & \textbf{0.0543} & 0.0868 & \textbf{0.0410} \\
|
||
\bottomrule
|
||
\end{tabularx}
|
||
\caption{Evaluation Metrics for the Ovulation-Over Target across Different Model Parameters for the Transformer model with
|
||
a fixed input window length of 160 days and an input resolution of 12 values per day.
|
||
\textbf{Bold} values represent the best value for each metric within a model.}
|
||
\label{tab:ov_over_results_by_model_parameters_transformer}
|
||
\end{table}
|
||
|
||
\begin{table}
|
||
\scriptsize
|
||
\begin{tabularx}{\linewidth}{l*{8}{X}}
|
||
\midrule
|
||
\multicolumn{8}{c}{\textbf{Convolutional-LSTM}} \\
|
||
\toprule
|
||
\multirow{2}{*}{Hidden Layer Size} &
|
||
\multirow{2}{*}{\# LSTM Layers} &
|
||
\multicolumn{3}{c}{MAE} &
|
||
\multicolumn{3}{c}{MSE} \\
|
||
\cmidrule(lr){3-5} \cmidrule(lr){6-8}
|
||
& & Fert Overall & Fert Days & Non-Fert Days & Fert Overall & Fert Days & Non-Fert Days \\
|
||
\midrule
|
||
16 & 1 & 0.2920 & 0.2198 & 0.3344 & 0.1264 & 0.0602 & 0.1653 \\
|
||
32 & 1 & 0.2141 & 0.1598 & 0.2377 & 0.0912 & 0.0398 & 0.1149 \\
|
||
32 & 2 & 0.1990 & 0.1556 & 0.2135 & 0.0852 & 0.0419 & 0.1025 \\
|
||
64 & 2 & 0.1743 & 0.1345 & 0.1845 & 0.0789 & 0.0400 & 0.0923 \\
|
||
128 & 2 & 0.1524 & 0.1138 & 0.1633 & 0.0717 & \textbf{0.0357} & 0.0845 \\
|
||
128 & 4 & 0.1579 & 0.1185 & 0.1670 & 0.0777 & 0.0401 & 0.0894 \\
|
||
256 & 4 & \textbf{0.1424} & 0.1425 & \textbf{0.1369} & \textbf{0.0687} & 0.0546 & \textbf{0.0715} \\
|
||
512 & 4 & 0.1436 & \textbf{0.1166} & 0.1523 & 0.0699 & 0.0382 & 0.0820 \\
|
||
\bottomrule
|
||
\end{tabularx}
|
||
\caption{Evaluation Metrics for the Ovulation-Over Target across Different Model Parameters for the convolutional LSTM model with
|
||
a fixed input window length of 40 days.
|
||
\textbf{Bold} values represent the best value for each metric within a model.}
|
||
\label{tab:ov_over_results_by_model_parameters_conv_lstm}
|
||
\end{table}
|
||
\begin{table}
|
||
\scriptsize
|
||
\begin{tabularx}{\linewidth}{l*{8}{X}}
|
||
\midrule
|
||
\multicolumn{8}{c}{\textbf{Convolutional-Transformer}} \\
|
||
\toprule
|
||
\multirow{2}{*}{Size of Embedding} &
|
||
\multirow{2}{*}{\# Encoder Layers} &
|
||
\multirow{2}{*}{\# Attention Heads} &
|
||
\multicolumn{3}{c}{MAE} &
|
||
\multicolumn{3}{c}{MSE} \\
|
||
\cmidrule(lr){4-6} \cmidrule(lr){7-9}
|
||
& & & Fert Overall & Fert Days & Non-Fert Days & Fert Overall & Fert Days & Non-Fert Days \\
|
||
\midrule
|
||
16 & 1 & 1 & 0.3759 & 0.3602 & 0.3855 & 0.1597 & 0.1498 & 0.1659 \\
|
||
32 & 1 & 1 & 0.1757 & 0.1119 & 0.2017 & 0.0916 & 0.0314 & 0.1167 \\
|
||
64 & 1 & 1 & 0.2771 & 0.2395 & 0.2978 & 0.1303 & 0.1133 & 0.1402 \\
|
||
64 & 2 & 2 & 0.1610 & 0.1149 & 0.1777 & 0.0816 & \textbf{0.0310} & 0.1021 \\
|
||
128 & 2 & 2 & 0.1501 & 0.0997 & 0.1725 & 0.0783 & 0.0316 & 0.0991 \\
|
||
128 & 4 & 4 & 0.1668 & 0.1472 & 0.1758 & 0.0801 & 0.0488 & 0.0953 \\
|
||
256 & 4 & 4 & 0.1529 & 0.1317 & 0.1597 & 0.0742 & 0.0435 & 0.0861 \\
|
||
512 & 4 & 4 & \textbf{0.1495} & 0.1298 & \textbf{0.1551} & \textbf{0.0703} & 0.0409 & \textbf{0.0814} \\
|
||
512 & 8 & 8 & 0.1576 & \textbf{0.1025} & 0.1848 & 0.0825 & 0.0339 & 0.1066 \\
|
||
\bottomrule
|
||
\end{tabularx}
|
||
\caption{Evaluation Metrics for the Ovulation-Over Target across Different Model Parameters for the convolutional Transformer model with
|
||
a fixed input window length of 40 days.
|
||
\textbf{Bold} values represent the best value for each metric within a model.}
|
||
\label{tab:ov_over_results_by_model_parameters_conv_transformer}
|
||
\end{table}
|
||
\end{landscape}
|
||
|
||
\subsubsection{Best Model Configuration Selection}\label{subsubsec:results_best_model_config_selection}
|
||
|
||
Following the selection procedure described in Section~\ref{subsubsec:methodology_best_model_config_selection}
|
||
the best configuration for each model architecture was identified based on the Fertility-Overall MSE and, where applicable,
|
||
the general tendencies of the model.
|
||
Table~\ref{tab:best_configs_lstm} and~\ref{tab:best_configs_transformer} summarize the selected input window length,
|
||
input resolution, and model complexity for each architecture.
|
||
These configurations are used in all subsequent experiments, including the irregular cycles analysis and the use case evaluation.
|
||
|
||
\begin{table}[htbp]
|
||
\centering
|
||
\scriptsize
|
||
\begin{tabularx}{\linewidth}{lXXXX}
|
||
\toprule
|
||
\textbf{Architecture} &
|
||
\textbf{Input Window Length} &
|
||
\textbf{Input Resolution} &
|
||
\textbf{Hidden Layer Size} &
|
||
\textbf{\# LSTM Layers} \\
|
||
\midrule
|
||
LSTM & 160 & 12 & 512 & 4 \\
|
||
Convolution-LSTM & 40 & 288 & 512 & 4 \\
|
||
\bottomrule
|
||
\end{tabularx}
|
||
\caption{Selected configurations for LSTM-based architectures. Input Window Length is given in days and input resolution in values per day.}
|
||
\label{tab:best_configs_lstm}
|
||
\end{table}
|
||
|
||
\begin{table}[htbp]
|
||
\centering
|
||
\scriptsize
|
||
\begin{tabularx}{\linewidth}{lXXXXX}
|
||
\toprule
|
||
\textbf{Architecture} &
|
||
\textbf{Input Window Length} &
|
||
\textbf{Input Resolution} &
|
||
\textbf{Embedding Size} &
|
||
\textbf{\# Encoder Layers} &
|
||
\textbf{\# Attention Heads} \\
|
||
\midrule
|
||
Transformer & 160 & 12 & 512 & 4 & 4 \\
|
||
Convolution-Transformer & 40 & 288 & 512 & 4 & 4 \\
|
||
\bottomrule
|
||
\end{tabularx}
|
||
\caption{Selected configurations for Transformer-based architectures. Input Window Length is given in days and input resolution in values per day.}
|
||
\label{tab:best_configs_transformer}
|
||
\end{table}
|
||
|
||
\subsection{Stratified Analysis}\label{subsec:stratified_analysis}
|
||
|
||
\subsubsection{Influence of User History Depth}\label{subsubsec:influence_of_past_user_data}
|
||
% don't forget to also add baseline to tables
|
||
|
||
\subsubsection{Regular vs Irregular Cycles}\label{subsubsec:regular_vs_irregular_cycles}
|
||
\begin{landscape}
|
||
\begin{table}
|
||
\small
|
||
\begin{tabularx}{\linewidth}{l*{6}{X}}
|
||
\toprule
|
||
\multirow{2}{*}{Model Architecture} & \multicolumn{3}{c}{MAE} & \multicolumn{3}{c}{MSE} \\
|
||
\cmidrule(r){2-4} \cmidrule(r){5-7}
|
||
& Fertility Overall & Fertile Days & Non-Fertile Days & Fertility Overall & Fertile Days & Non-Fertile Days \\
|
||
\midrule
|
||
\multicolumn{7}{c}{\textbf{Regular Cycle Group}} \\
|
||
\midrule
|
||
|
||
\midrule
|
||
\multicolumn{7}{c}{\textbf{Irregular Cycle Group}} \\
|
||
\midrule
|
||
|
||
\bottomrule
|
||
\end{tabularx}
|
||
\caption{Evaluation Metrics for the Fertility-Probability Target across Different Model Architectures for the Regular and Irregular Cycle Groups.
|
||
\underline{Underlined} values represent the best value for each metric within a model.
|
||
\textbf{\underline{Bold + Underlined}} values represent the global best values across all models for a given metric.}
|
||
\label{tab:regular_vs_irregular_fertility_results}
|
||
\end{table}
|
||
\end{landscape}
|
||
\begin{landscape}
|
||
\begin{table}
|
||
\small
|
||
\begin{tabularx}{\linewidth}{l*{6}{X}}
|
||
\toprule
|
||
\multirow{2}{*}{Model} & \multicolumn{3}{c}{MAE} & \multicolumn{3}{c}{MSE} \\
|
||
\cmidrule(r){2-4} \cmidrule(r){5-7}
|
||
& OV-Over Overall & OV-Over Before OV & OV-Over After OV & OV-Over Overall & OV-Over Before OV & OV-Over After OV \\
|
||
\midrule
|
||
\multicolumn{7}{c}{\textbf{Regular Cycle Group}} \\
|
||
\midrule
|
||
|
||
\midrule
|
||
\multicolumn{7}{c}{\textbf{Irregular Cycle Group}} \\
|
||
\midrule
|
||
|
||
\bottomrule
|
||
\end{tabularx}
|
||
\caption{Evaluation Metrics for the Ovulation-Over Target across Different Model Architectures for the Regular and Irregular Cycle Groups.
|
||
\underline{Underlined} values represent the best value for each metric within a model.
|
||
\textbf{\underline{Bold + Underlined}} values represent the global best values across all models for a given metric.}
|
||
\label{tab:regular_vs_irregular_ov_over_results}
|
||
\end{table}
|
||
\end{landscape}
|
||
|
||
\subsection{Use-Case Evaluation Results}\label{subsec:use_case_evaluation_results}
|
||
|
||
\subsubsection{Contraception Use-Case Results}\label{subsubsec:use_case_contraception_results}
|
||
|
||
\subsubsection{Pregnancy Use-Case Results}\label{subsubsec:use_case_pregnancy_results}
|
||
|
||
\subsection{Summary of Key Findings}\label{subsec:summary_of_key_findings}
|